The FREQ Procedure

TABLES Statement

Subsections:

Without Options
Options

TABLES requests < / options > ;

The TABLES statement requests one-way to n-way frequency and crosstabulation tables and statistics for those tables.

If you omit the TABLES statement, PROC FREQ generates one-way frequency tables for all data set variables that are not listed in the other statements.

The following argument is required in the TABLES statement.

requests

specify the frequency and crosstabulation tables to produce. A request is composed of one variable name or several variable names separated by asterisks. To request a one-way frequency table, use a single variable. To request a two-way crosstabulation table, use an asterisk between two variables. To request a multiway table (an n-way table, where n>2), separate the desired variables with asterisks. The unique values of these variables form the rows, columns, and strata of the table. You can include up to 50 variables in a single multiway table request.

For two-way to multiway tables, the values of the last variable form the crosstabulation table columns, while the values of the next-to-last variable form the rows. Each level (or combination of levels) of the other variables forms one stratum. PROC FREQ produces a separate crosstabulation table for each stratum. For example, a specification of A*B*C*D in a TABLES statement produces k tables, where k is the number of different combinations of values for A and B. Each table lists the values for C down the side and the values for D across the top.

You can use multiple TABLES statements in the PROC FREQ step. PROC FREQ builds all the table requests in one pass of the data, so that there is essentially no loss of efficiency. You can also specify any number of table requests in a single TABLES statement. To specify multiple table requests quickly, use a grouping syntax by placing parentheses around several variables and joining other variables or variable combinations. For example, the statements shown in Table 3.8 illustrate grouping syntax.

Table 3.8: Grouping Syntax

TABLES Request	Equivalent to
`A`*(`B C`)	`A``B` `A``C`
(`A B`)*(`C D`)	`A``C` `B``C` `A``D` `B``D`
(`A B C`)*`D`	`A``D` `B``D` `C`*`D`
`A` – – `C`	`A` `B` `C`
(`A` – – `C`)*`D`	`A``D` `B``D` `C`*`D`

The TABLES statement variables are one or more variables from the DATA= input data set. These variables can be either character or numeric, but the procedure treats them as categorical variables. PROC FREQ uses the formatted values of the TABLES variable to determine the categorical variable levels. So if you assign a format to a variable with a FORMAT statement, PROC FREQ formats the values before dividing observations into the levels of a frequency or crosstabulation table. See the FORMAT procedure in the Base SAS Procedures Guide and the FORMAT statement and SAS formats in SAS Formats and Informats: Reference.

If you use PROC FORMAT to create a user-written format that combines missing and nonmissing values into one category, PROC FREQ treats the entire category of formatted values as missing. See the discussion in the section Grouping with Formats for more information.

By default, the frequency or crosstabulation table lists the values of both character and numeric variables in ascending order based on internal (unformatted) variable values. You can change the order of the values in the table by specifying the ORDER= option in the PROC FREQ statement. To list the values in ascending order by formatted value, use ORDER=FORMATTED.

Without Options

If you request a one-way frequency table for a variable without specifying options, PROC FREQ produces frequencies, cumulative frequencies, percentages of the total frequency, and cumulative percentages for each value of the variable. If you request a two-way or an n-way crosstabulation table without specifying options, PROC FREQ produces crosstabulation tables that include cell frequencies, cell percentages of the total frequency, cell percentages of row frequencies, and cell percentages of column frequencies. The procedure excludes observations with missing values from the table but displays the total frequency of missing observations below each table.

Options

Table 3.9 lists the options available in the TABLES statement. Descriptions of the options follow in alphabetical order.

Table 3.9: TABLES Statement Options

Option	Description
Control Statistical Analysis
AGREE	Requests tests and measures of classification agreement
ALL	Requests tests and measures of association produced by the
	CHISQ, MEASURES, and CMH options
ALPHA=	Sets confidence level for confidence limits
BINOMIAL \| BIN	Requests binomial proportions, confidence limits, and tests
	for one-way tables
CHISQ	Requests chi-square tests and measures based on chi-square
CL	Requests confidence limits for MEASURES statistics
CMH	Requests all Cochran-Mantel-Haenszel statistics
CMH1	Requests CMH correlation statistic, adjusted odds ratios,
	and adjusted relative risks
CMH2	Requests CMH correlation and row mean scores (ANOVA)
	statistics, adjusted odds ratios, and adjusted relative risks
FISHER	Requests Fisher’s exact test for tables larger than $2 \times 2$
GAILSIMON	Requests Gail-Simon test for qualitative interactions
JT	Requests Jonckheere-Terpstra test
MEASURES	Requests measures of association
MISSING	Treats missing values as nonmissing
PLCORR	Requests polychoric correlation
RELRISK	Requests odds ratio and relative risks for $2 \times 2$ tables
RISKDIFF	Requests risks and risk differences for $2 \times 2$ tables
SCORES=	Specifies type of row and column scores
TREND	Requests Cochran-Armitage test for trend
Control Additional Table Information
CELLCHI2	Displays cell contributions to the Pearson chi-square statistic
CUMCOL	Displays cumulative column percentages
DEVIATION	Displays deviations of cell frequencies from expected values
EXPECTED	Displays expected cell frequencies
MISSPRINT	Displays missing value frequencies
PEARSONRES	Displays Pearson residuals in the CROSSLIST table
PRINTKWTS	Displays kappa coefficient weights
SCOROUT	Displays row and column scores
SPARSE	Includes all possible combinations of variable levels in the
	LIST table and OUT= data set
STDRES	Displays standardized residuals in the CROSSLIST table
TOTPCT	Displays percentages of total frequency for n-way tables (n>2)
Control Displayed Output
CONTENTS=	Specifies contents label for crosstabulation tables
CROSSLIST	Displays crosstabulation tables in ODS column format
FORMAT=	Formats frequencies in crosstabulation tables
LIST	Displays two-way to n-way tables in list format
MAXLEVELS=	Specifies maximum number of levels to display in one-way tables
NOCOL	Suppresses display of column percentages
NOCUM	Suppresses display of cumulative frequencies and percentages
NOFREQ	Suppresses display of frequencies
NOPERCENT	Suppresses display of percentages
NOPRINT	Suppresses display of crosstabulation tables but displays statistics
NOROW	Suppresses display of row percentages
NOSPARSE	Suppresses zero frequency levels in the CROSSLIST table,
	LIST table, and OUT= data set
NOWARN	Suppresses log warning message for the chi-square test
Produce Statistical Graphics
PLOTS=	Requests plots from ODS Graphics
Create an Output Data Set
OUT=	Names an output data set to contain frequency counts
OUTCUM	Includes cumulative frequencies and percentages in the
	output data set for one-way tables
OUTEXPECT	Includes expected frequencies in the output data set
OUTPCT	Includes row, column, and two-way table percentages in the
	output data set

You can specify the following options in a TABLES statement.

AGREE < (options)>

requests tests and measures of classification agreement for square tables. This option provides the simple and weighted kappa coefficients along with their standard errors and confidence limits. For multiway tables, the AGREE option also produces the overall simple and weighted kappa coefficients (along with their standard errors and confidence limits) and tests for equal kappas among strata. For $2 \times 2$ tables, this option provides McNemar’s test; for square tables that have more than two response categories (levels), this option provides Bowker’s test of symmetry. For multiway tables that have two response categories, the AGREE option also produces Cochran’s Q test. For more information, see the section Tests and Measures of Agreement.

Measures of agreement can be computed only for square tables, where the number of rows equals the number of columns. If your table is not square because of observations that have zero weights, you can specify the ZEROS option in the WEIGHT statement to include these observations in the analysis. For more information, see the section Tables with Zero Rows and Columns.

You can set the level for agreement confidence limits by specifying the ALPHA= option in the TABLES statement. The default of ALPHA=0.05 produces 95% confidence limits.

You can specify the TEST statement to request asymptotic tests for the simple and weighted kappa coefficients. You can specify the EXACT statement to request McNemar’s exact test (for $2 \times 2$ tables) and exact tests for the simple and weighted kappa coefficients. For more information, see the section Exact Statistics.

The weighted kappa coefficient is computed by using agreement weights that reflect the relative agreement between pairs of variable levels. To specify the type of agreement weights and to display the agreement weights, you can specify the following options:

PRINTKWTS: displays the agreement weights that PROC FREQ uses to compute the weighted kappa coefficient. Agreement weights reflect the relative agreement between pairs of variable levels. By default, PROC FREQ uses the Cicchetti-Allison form of agreement weights. If you specify the WT=FC option, the procedure uses the Fleiss-Cohen form of agreement weights. For more information, see the section Weighted Kappa Coefficient.
WT=FC: requests Fleiss-Cohen agreement weights in the weighted kappa computation. By default, PROC FREQ uses Cicchetti-Allison agreement weights to compute the weighted kappa coefficient. Agreement weights reflect the relative agreement between pairs of variable levels. For more information, see the section Weighted Kappa Coefficient.

ALL

requests all tests and measures that are produced by the CHISQ, MEASURES, and CMH options. You can control the number of CMH statistics to compute by specifying the CMH1 or CMH2 option.

ALPHA= $\alpha$

specifies the level of confidence limits. The value of $\alpha$ must be between 0 and 1, and the default is 0.05. A confidence level of $\alpha$ produces $100(1-\alpha )$ % confidence limits. The default of ALPHA=0.05 produces 95% confidence limits.

ALPHA= applies to confidence limits that are requested by TABLES statement options. There is a separate ALPHA= option in the EXACT statement that sets the level of confidence limits for Monte Carlo estimates of exact p-values, which you request by specifying the MC option in the EXACT statement.

BINOMIAL < (binomial-options)> BIN <(binomial-options)>

requests the binomial proportion for one-way tables. When you specify the BINOMIAL option, by default PROC FREQ also provides the asymptotic standard error, asymptotic Wald and exact (Clopper-Pearson) confidence limits, and the asymptotic equality test for the binomial proportion.

You can specify binomial-options in parentheses after the BINOMIAL option. The LEVEL= binomial-option identifies the variable level for which to compute the proportion. If you do not specify LEVEL=, PROC FREQ computes the proportion for the first level that appears in the output. The P= binomial-option specifies the null proportion for the binomial tests. If you do not specify P=, PROC FREQ uses P=0.5 by default.

You can also specify binomial-options to request additional tests and confidence limits for the binomial proportion. The EQUIV, NONINF, and SUP binomial-options request tests of equivalence, noninferiority, and superiority, respectively. The CL= binomial-option requests confidence limits for the binomial proportion. Table 3.10 summarizes the binomial-options.

Available confidence limits for the binomial proportion include Agresti-Coull, exact (Clopper-Pearson), Jeffreys, Wald, and Wilson (score) confidence limits. You can specify more than one type of binomial confidence limits in the same analysis. If you do not specify any confidence limit requests, PROC FREQ computes Wald asymptotic confidence limits and exact (Clopper-Pearson) confidence limits by default. The ALPHA= option determines the confidence level, and the default of ALPHA=0.05 produces 95% confidence limits for the binomial proportion.

As part of the noninferiority, superiority, and equivalence analyses, PROC FREQ provides test-based confidence limits that have a confidence coefficient of $100(1-2\alpha )$ % (Schuirmann, 1999). The ALPHA= option determines the confidence level, and the default of ALPHA=0.05 produces 90% confidence limits. See the sections Noninferiority Test and Equivalence Test for details.

To request exact tests for the binomial proportion, specify the BINOMIAL option in the EXACT statement. PROC FREQ then computes exact p-values for all binomial tests that you request with binomial-options, which can include tests of noninferiority, superiority, and equivalence, in addition to the test of equality.

See the section Binomial Proportion for details.

Table 3.10: BINOMIAL Options

Option	Description
LEVEL=	Specifies the variable level
P=	Specifies the null proportion
CORRECT	Requests continuity correction
Request Confidence Limits
CL=AGRESTICOULL \| AC	Requests Agresti-Coull confidence limits
CL=ALL	Requests all confidence limits
CL=EXACT \| CLOPPERPEARSON	Requests Clopper-Pearson confidence limits
CL=JEFFREYS	Requests Jeffreys confidence limits
CL=WALD	Requests Wald confidence limits
CL=WILSON \| SCORE	Requests Wilson (score) confidence limits
Request Tests
EQUIV \| EQUIVALENCE	Requests an equivalence test
NONINF \| NONINFERIORITY	Requests a noninferiority test
SUP \| SUPERIORITY	Requests a superiority test
MARGIN=	Specifies the test margin
VAR=SAMPLE \| NULL	Specifies the test variance

You can specify the following binomial-options:

CL=type | (types)

requests confidence limits for the binomial proportion. You can specify one or more types of confidence limits. When you specify only one type, you can omit the parentheses around the request.

PROC FREQ displays the confidence limits in the “Binomial Confidence Limits” table. The ALPHA= option determines the level of the confidence limits that the CL= binomial-option provides. The default of ALPHA=0.05 produces 95% confidence limits for the binomial proportion. This differs from the test-based confidence limits that are provided with the equivalence, noninferiority, and superiority tests (EQUIV, NONINF, and SUP), which have a confidence coefficient of $100(1-2\alpha )$ % (Schuirmann, 1999). See the sections Equivalence Test and Noninferiority Test for details.

You can specify the CL= binomial-option with or without requests for binomial tests. The confidence limits that CL= produces do not depend on the tests that you request and do not use the value of the test margin (which you specify by using the MARGIN= binomial-option).

The following types of binomial confidence limits are available:

AGRESTICOULL AC

requests Agresti-Coull confidence limits for the binomial proportion. See the section Agresti-Coull Confidence Limits for details.

ALL

requests all available types of confidence limits for the binomial proportion. These include the following: Agresti-Coull, exact (Clopper-Pearson), Jeffreys, Wald, and Wilson (score) confidence limits.

EXACT CLOPPERPEARSON

requests exact (Clopper-Pearson) confidence limits for the binomial proportion. See the section Exact (Clopper-Pearson) Confidence Limits for details. If you do not request any binomial confidence limits by specifying the CL= binomial-option, PROC FREQ produces Wald and exact (Clopper-Pearson) confidence limits by default. To request exact tests for the binomial proportion, specify the BINOMIAL option in the EXACT statement.

JEFFREYS

requests Jeffreys confidence limits for the binomial proportion. See the section Jeffreys Confidence Limits for details.

WALD < (CORRECT)>

requests Wald confidence limits for the binomial proportion. See the section Wald Confidence Limits for details. If you do not request any binomial confidence limits by specifying the CL= binomial-option, PROC FREQ produces Wald and exact (Clopper-Pearson) confidence limits by default.

If you specify CL=WALD(CORRECT), the Wald confidence limits include a continuity correction. If you specify the CORRECT binomial-option, the Wald confidence limits and the Wald tests include continuity corrections.

WILSON < (CORRECT)> SCORE <(CORRECT)>

requests Wilson confidence limits for the binomial proportion. These are also known as score confidence limits. See the section Wilson (Score) Confidence Limits for details.

If you specify CL=WILSON(CORRECT) or the CORRECT binomial-option, the Wilson confidence limits include a continuity correction.

CORRECT

includes a continuity correction in the Wald confidence limits, Wald tests, and Wilson confidence limits.

You can request continuity corrections individually for Wald or Wilson confidence limits by specifying the CL=WALD(CORRECT) or CL=WILSON(CORRECT) binomial-option, respectively.

EQUIV EQUIVALENCE

requests a test of equivalence for the binomial proportion. See the section Equivalence Test for details. You can specify the equivalence test margins, the null proportion, and the variance type with the MARGIN=, P=, and VAR= binomial-options, respectively. To request an exact equivalence test, specify the BINOMIAL option in the EXACT statement.

LEVEL=level-number | 'level-value'

specifies the variable level for the binomial proportion. By default, PROC FREQ computes the proportion of observations for the first variable level that appears in the output. To request a different level, use LEVEL=level-number or LEVEL='level-value', where level-number is the variable level’s number or order in the output, and level-value is the formatted value of the variable level. The value of level-number must be a positive integer. You must enclose level-value in single quotes.

MARGIN=value | (lower, upper)

specifies the margin for the noninferiority, superiority, and equivalence tests, which you request with the NONINF, SUP, and EQUIV binomial-options, respectively. If you do not specify this option, PROC FREQ uses a margin of 0.2 by default.

For noninferiority and superiority tests, specify a single value for the MARGIN= option. The MARGIN= value must be a positive number. You can specify value as a number between 0 and 1. Or you can specify value in percentage form as a number between 1 and 100, and PROC FREQ converts that number to a proportion. The procedure treats the value 1 as 1%.

For noninferiority and superiority tests, the test limits must be between 0 and 1. The limits are determined by the null proportion value (which you can specify with the P= binomial-option) and by the margin value. The noninferiority limit equals the null proportion minus the margin. By default, the null proportion equals 0.5 and the margin equals 0.2, which gives a noninferiority limit of 0.3. The superiority limit equals the null proportion plus the margin, which is 0.7 by default.

For an equivalence test, you can specify a single MARGIN= value, or you can specify both lower and upper values. If you specify a single MARGIN= value, it must be a positive number, as described previously. If you specify a single MARGIN= value for an equivalence test, PROC FREQ uses –value as the lower margin and value as the upper margin for the test. If you specify both lower and upper values for an equivalence test, you can specify them in proportion form as numbers between –1 or 1. Or you can specify them in percentage form as numbers between –100 and 100, and PROC FREQ converts the numbers to proportions. The value of lower must be less than the value of upper.

The equivalence limits must be between 0 and 1. The equivalence limits are determined by the null proportion value (which you can specify with the P= binomial-option) and by the margin values. The lower equivalence limit equals the null proportion plus the lower margin. By default, the null proportion equals 0.5 and the lower margin equals –0.2, which gives a lower equivalence limit of 0.3. The upper equivalence limit equals the null proportion plus the upper margin, which is 0.7 by default.

See the sections Noninferiority Test and Equivalence Test for details.

NONINF NONINFERIORITY

requests a test of noninferiority for the binomial proportion. See the section Noninferiority Test for details. You can specify the noninferiority test margin, the null proportion, and the variance type with the MARGIN=, P=, and VAR= binomial-options, respectively. To request an exact noninferiority test, specify the BINOMIAL option in the EXACT statement.

P=value

specifies the null hypothesis proportion for the binomial tests. If you omit this option, PROC FREQ uses 0.5 as the null proportion. The null proportion value must be a positive number. You can specify value as a number between 0 and 1. Or you can specify value in percentage form as a number between 1 and 100, and PROC FREQ converts that number to a proportion. The procedure treats the value 1 as 1%.

SUP SUPERIORITY

requests a test of superiority for the binomial proportion. See the section Superiority Test for details. You can specify the superiority test margin, the null proportion, and the variance type with the MARGIN=, P=, and VAR= binomial-options, respectively. To request an exact superiority test, specify the BINOMIAL option in the EXACT statement.

VAR=SAMPLE | NULL

specifies the type of variance estimate to use in the Wald tests of noninferiority, superiority, and equivalence. The default is VAR=SAMPLE, which estimates the variance from the sample proportion. VAR=NULL uses a test-based variance that is computed from the null hypothesis proportion (which is specified by the P= binomial-option). See the sections Noninferiority Test and Equivalence Test for details.

CELLCHI2

displays each table cell’s contribution to the Pearson chi-square statistic in the crosstabulation table. The cell chi-square is computed as $(\emph{frequency}-\emph{expected})^2/\emph{expected}$ , where frequency is the table cell frequency (count) and expected is the expected cell frequency, which is computed under the null hypothesis that the row and column variables are independent. For more information, see the section Pearson Chi-Square Test for Two-Way Tables. This option has no effect for one-way tables or for tables that are displayed in list format (which you can request by specifying the LIST option).

CHISQ < (chisq-options)>

requests chi-square tests of homogeneity or independence and measures of association that are based on the chi-square statistic. For two-way tables, the chi-square tests include the Pearson chi-square, likelihood ratio chi-square, and Mantel-Haenszel chi-square tests. The chi-square measures include the phi coefficient, contingency coefficient, and Cramér’s V. For $2 \times 2$ tables, the CHISQ option also provides Fisher’s exact test and the continuity-adjusted chi-square test. See the section Chi-Square Tests and Statistics for details.

For one-way tables, the CHISQ option provides the Pearson chi-square goodness-of-fit test. You can also request the likelihood ratio goodness-of-fit test for one-way tables by specifying the LRCHI chisq-option in parentheses after the CHISQ option. By default, the one-way chi-square tests are based on the null hypothesis of equal proportions. Alternatively, you can provide null hypothesis proportions or frequencies by specifying the TESTP= or TESTF= chisq-option, respectively. See the section Chi-Square Test for One-Way Tables for more information.

To request Fisher’s exact test for tables larger than $2 \times 2$ , specify the FISHER option in the EXACT statement. Exact p-values are also available for the Pearson, likelihood ratio, and Mantel-Haenszel chi-square tests. See the description of the EXACT statement for more information.

You can specify the following chisq-options:

DF=df

specifies the degrees of freedom for the chi-square tests. The value of df must not be zero. If the value of df is positive, PROC FREQ uses df as the degrees of freedom for the chi-square tests. If the value of df is negative, PROC FREQ uses df to adjust the default degrees of freedom for the chi-square tests.

By default for one-way tables, the value of df is (n – 1), where n is the number of variable levels in the table. By default for two-way tables, the value of df is (r – 1) (c – 1), where r is the number of rows in the table and c is the number of columns. See the sections Chi-Square Test for One-Way Tables and Chi-Square Tests and Statistics for more information.

If you specify a negative value of df, PROC FREQ adjusts the default degrees of freedom by adding the (negative) value of df to the default value to produce the adjusted degrees of freedom. The adjusted degrees of freedom must be positive.

The DF= chisq-option specifies or adjusts the degrees of freedom for the following chi-square tests: the Pearson and likelihood ratio goodness-of-fit tests for one-way tables; and the Pearson, likelihood ratio, and Mantel-Haenszel chi-square tests for two-way tables.

LRCHI

requests the likelihood ratio goodness-of-fit test for one-way tables. See the section Likelihood Ratio Chi-Square Test for One-Way Tables for more information.

By default, this test is based on the null hypothesis of equal proportions. You can provide null hypothesis proportions or frequencies by specifying the TESTP= or TESTF= chisq-option, respectively. You can request an exact likelihood ratio goodness-of-fit test by specifying the LRCHI option in the EXACT statement.

TESTF=(values)| SAS-data-set

specifies null hypothesis frequencies for the one-way chi-square goodness-of-fit tests. See the section Chi-Square Test for One-Way Tables for details. You can list the null frequencies as values in parentheses after TESTF=. Or you can provide the null frequencies in a secondary input data set by specifying TESTF=SAS-data-set. The TESTF=SAS-data-set cannot be the same data set that you specify in the DATA= option. You can specify only one TESTF= or TESTP= data set in a single invocation of the procedure.

If you list the null frequencies as values, you can separate the values with blanks or commas. The values must be positive numbers. The number of values must equal the number of variable levels in the one-way table. The sum of the values must equal the total frequency for the one-way table. Order the values to match the order in which the corresponding variable levels appear in the one-way frequency table.

If you provide the null frequencies in a secondary input data set (TESTF=SAS-data-set), the variable that contains the null frequencies should be named _TESTF_, TestFrequency, or Frequency. The null frequencies must be positive numbers. The number of frequencies must equal the number of levels in the one-way frequency table, and the sum of the frequencies must equal the total frequency for the one-way table. Order the null frequencies in the data set to match the order in which the corresponding variable levels appear in the one-way frequency table.

TESTP=(values)| SAS-data-set

specifies null hypothesis proportions for the one-way chi-square goodness-of-fit tests. See the section Chi-Square Test for One-Way Tables for details. You can list the null proportions as values in parentheses after TESTP=. Or you can provide the null proportions in a secondary input data set by specifying TESTP=SAS-data-set. The TESTP=SAS-data-set cannot be the same data set that you specify in the DATA= option. You can specify only one TESTF= or TESTP= data set in a single invocation of the procedure.

If you list the null proportions as values, you can separate the values with blanks or commas. The values must be positive numbers. The number of values must equal the number of variable levels in the one-way table. Order the values to match the order in which the corresponding variable levels appear in the one-way frequency table. You can specify values in probability form as numbers between 0 and 1, where the proportions sum to 1. Or you can specify values in percentage form as numbers between 0 and 100, where the percentages sum to 100.

If you provide the null proportions in a secondary input data set (TESTP=SAS-data-set), the variable that contains the null proportions should be named _TESTP_, TestPercent, or Percent. The null proportions must be positive numbers. The number of proportions must equal the number of levels in the one-way frequency table. You can provide the proportions in probability form as numbers between 0 and 1, where the proportions sum to 1. Or you can provide the proportions in percentage form as numbers between 0 and 100, where the percentages sum to 100. Order the null proportions in the data set to match the order in which the corresponding variable levels appear in the one-way frequency table.

WARN=type | (types)

controls the warning message for the validity of the asymptotic Pearson chi-square test. By default, PROC FREQ displays a warning message when more than 20% of the table cells have expected frequencies that are less than 5. If you specify the NOPRINT option in the PROC FREQ statement, the procedure displays the warning in the log; otherwise, the procedure displays the warning as a footnote in the chi-square table. You can use the WARN= option to suppress the warning and to include a warning indicator in the output data set.

You can specify one or more of the following types in the WARN= option. If you specify more than one type value, enclose the values in parentheses after WARN=. For example, warn = (output noprint).

Value of WARN=	Description
OUTPUT	Adds a warning indicator variable to the output data set
NOLOG	Suppresses the chi-square warning message in the log
NOPRINT	Suppresses the chi-square warning message in the display
NONE	Suppresses the chi-square warning message entirely

If you specify the WARN=OUTPUT option, the ODS output data set ChiSq contains a variable named Warning that equals 1 for the Pearson chi-square observation when more than 20% of the table cells have expected frequencies that are less than 5 and equals 0 otherwise. If you specify WARN=OUTPUT and also specify the CHISQ option in the OUTPUT statement, the statistics output data set contains a variable named WARN_PCHI that indicates the warning.

The WARN=NOLOG option has the same effect as the NOWARN option in the TABLES statement.

CL

requests confidence limits for the measures of association, which you can request by specifying the MEASURES option. For more information, see the sections Measures of Association and Confidence Limits. You can set the level of the confidence limits by using the ALPHA= option. The default of ALPHA=0.05 produces 95% confidence limits.

If you omit the MEASURES option, the CL option invokes MEASURES. The CL option is equivalent to the MEASURES(CL) option.

CMH < (cmh-options)>

requests Cochran-Mantel-Haenszel statistics, which test for association between the row and column variables after adjusting for the remaining variables in a multiway table. The Cochran-Mantel-Haenszel statistics include the nonzero correlation statistic, the row mean scores (ANOVA) statistic, and the general association statistic. In addition, for $2 \times 2$ tables, the CMH option provides the adjusted Mantel-Haenszel and logit estimates of the odds ratio and relative risks, together with their confidence limits. For stratified $2 \times 2$ tables, the CMH option provides the Breslow-Day test for homogeneity of odds ratios. (To request Tarone’s adjustment for the Breslow-Day test, specify the BDT cmh-option.) See the section Cochran-Mantel-Haenszel Statistics for details.

You can use the CMH1 or CMH2 option to control the number of CMH statistics that PROC FREQ computes.

For stratified $2 \times 2$ tables, you can request Zelen’s exact test for equal odds ratios by specifying the EQOR option in the EXACT statement. See the section Zelen’s Exact Test for Equal Odds Ratios for details. You can request exact confidence limits for the common odds ratio by specifying the COMOR option in the EXACT statement. This option also provides a common odds ratio test. See the section Exact Confidence Limits for the Common Odds Ratio for details.

You can specify the following cmh-options in parentheses after the CMH option. These cmh-options, which apply to stratified $2 \times 2$ tables, are also available with the CMH1 or CMH2 option.

BDT

requests Tarone’s adjustment in the Breslow-Day test for homogeneity of odds ratios. See the section Breslow-Day Test for Homogeneity of the Odds Ratios for details.

GAILSIMON < (COLUMN=1 | 2)> GS < (COLUMN=1 | 2)>

requests the Gail-Simon test for qualitative interaction, which applies to stratified $2 \times 2$ tables. See the section Gail-Simon Test for Qualitative Interactions for details.

The COLUMN= option specifies the column of the risk differences to use in computing the Gail-Simon test. By default, PROC FREQ uses column 1 risk differences. If you specify COLUMN=2, PROC FREQ uses column 2 risk differences.

The GAILSIMON cmh-option has the same effect as the GAILSIMON option in the TABLES statement.

MANTELFLEISS MF

requests the Mantel-Fleiss criterion for the Mantel-Haenszel statistic for stratified $2 \times 2$ tables. See the section Mantel-Fleiss Criterion for details.

CMH1 < (cmh-options)>

requests the Cochran-Mantel-Haenszel correlation statistic. This option does not provide the CMH row mean scores (ANOVA) statistic or the general association statistic, which are provided by the CMH option. For tables larger than $2 \times 2$ , the CMH1 option requires less memory than the CMH option, which can require an enormous amount of memory for large tables.

For $2 \times 2$ tables, the CMH1 option also provides the adjusted Mantel-Haenszel and logit estimates of the odds ratio and relative risks, together with their confidence limits. For stratified $2 \times 2$ tables, the CMH1 option provides the Breslow-Day test for homogeneity of odds ratios.

The cmh-options for CMH1 are the same as the cmh-options that are available with the CMH option. See the description of the CMH option for details.

CMH2 < (cmh-options)>

requests the Cochran-Mantel-Haenszel correlation statistic and the row mean scores (ANOVA) statistic. This option does not provide the CMH general association statistic, which is provided by the CMH option. For tables larger than $2 \times 2$ , the CMH2 option requires less memory than the CMH option, which can require an enormous amount of memory for large tables.

The cmh-options for CMH2 are the same as the cmh-options that are available with the CMH option. See the description of the CMH option for details.

CONTENTS='string'

specifies the label to use for crosstabulation tables in the contents file, the Results window, and the trace record. For information about output presentation, see the SAS Output Delivery System: User's Guide.

If you omit the CONTENTS= option, the contents label for crosstabulation tables is “Cross-Tabular Freq Table” by default.

Note that contents labels for all crosstabulation tables that are produced by a single TABLES statement use the same text. To specify different contents labels for different crosstabulation tables, request the tables in separate TABLES statements and use the CONTENTS= option in each TABLES statement.

To remove the crosstabulation table entry from the contents file, you can specify a null label with CONTENTS=''.

The CONTENTS= option affects only contents labels for crosstabulation tables. It does not affect contents labels for other PROC FREQ tables.

To specify the contents label for any PROC FREQ table, you can use PROC TEMPLATE to create a customized table template. The CONTENTS_LABEL attribute in the DEFINE TABLE statement of PROC TEMPLATE specifies the contents label for the table. See the chapter “The TEMPLATE Procedure” in the SAS Output Delivery System: User's Guide for more information.

CROSSLIST < (options)>

displays crosstabulation tables by using an ODS column format instead of the default crosstabulation cell format. In the CROSSLIST table display, the rows correspond to the crosstabulation table cells, and the columns correspond to descriptive statistics such as frequencies and percentages. The CROSSLIST table displays the same information as the default crosstabulation table (but it uses an ODS column format). For more information about the contents of the CROSSLIST table, See the section Two-Way and Multiway Tables.

You can control the contents of a CROSSLIST table by specifying the same options available for the default crosstabulation table. These include the NOFREQ, NOPERCENT, NOROW, and NOCOL options. You can request additional information in a CROSSLIST table by specifying the CELLCHI2, DEVIATION, EXPECTED, MISSPRINT, and TOTPCT options. You can also display standardized residuals or Pearson residuals in a CROSSLIST table by specifying the CROSSLIST(STDRES) or CROSSLIST(PEARSONRES) option, respectively; these options are not available for the default crosstabulation table. The FORMAT= and CUMCOL options have no effect on CROSSLIST tables. You cannot specify both the LIST option and the CROSSLIST option in the same TABLES statement.

For CROSSLIST tables, you can use the NOSPARSE option to suppress display of variable levels that have zero frequencies. By default, PROC FREQ displays all levels of the column variable within each level of the row variable, including any levels that have zero frequencies. By default for multiway tables that are displayed as CROSSLIST tables, the procedure displays all levels of the row variable for each stratum of the table, including any row levels with zero frequencies in the stratum.

You can specify the following options:

STDRES: displays the standardized residuals of the table cells in the CROSSLIST table. The standardized residual is the ratio of (frequency – expected) to its standard error, where frequency is the table cell frequency (count) and expected is the expected table cell frequency, which is computed under the null hypothesis that the row and column variables are independent. For more information, see the section Standardized Residuals. You can display the expected values and deviations by specifying the EXPECTED and DEVIATION options, respectively.
PEARSONRES: displays the Pearson residuals of the table cells in the CROSSLIST table. The Pearson residual is the square root of the table cell’s contribution to the Pearson chi-square statistic. The Pearson residual is computed as $(\emph{frequency}-\emph{expected} )/\sqrt {\emph{expected}}$ , where frequency is the table cell frequency (count) and expected is the expected table cell frequency, which is computed under the null hypothesis that the row and column variables are independent. For more information, see the section Pearson Chi-Square Test for Two-Way Tables. You can display the expected values, deviations, and cell chi-squares by specifying the EXPECTED, DEVIATION, and CELLCHI2 options, respectively.

CUMCOL

displays the cumulative column percentages in the cells of the crosstabulation table. The CUMCOL option does not apply to crosstabulation tables produced with the LIST or CROSSLIST option.

DEVIATION

displays the deviations of the frequencies from the expected frequencies (frequency – expected) in the crosstabulation table. The expected frequencies are computed under the null hypothesis that the row and column variables are independent. For more information, see the section Pearson Chi-Square Test for Two-Way Tables. You can display the expected values by specifying the EXPECTED option. This option has no effect for one-way tables or for tables that are displayed in list format (which you can request by specifying the LIST option).

EXPECTED

displays the expected cell frequencies in the crosstabulation table. The expected frequencies are computed under the null hypothesis that the row and column variables are independent. For more information, see the section Pearson Chi-Square Test for Two-Way Tables. This option has no effect for one-way tables or for tables that are displayed in list format (which you can request by specifying the LIST option).

FISHER

requests Fisher’s exact test for tables that are larger than $2 \times 2$ . (For $2 \times 2$ tables, the CHISQ option provides Fisher’s exact test.) This test is also known as the Freeman-Halton test. See the sections Fisher’s Exact Test and Exact Statistics for more information.

If you omit the CHISQ option in the TABLES statement, the FISHER option invokes CHISQ. You can also request Fisher’s exact test by specifying the FISHER option in the EXACT statement.

Note: PROC FREQ computes exact tests by using fast and efficient algorithms that are superior to direct enumeration. Exact tests are appropriate when a data set is small, sparse, skewed, or heavily tied. For some large problems, computation of exact tests might require a substantial amount of time and memory. Consider using asymptotic tests for such problems. Alternatively, when asymptotic methods might not be sufficient for such large problems, consider using Monte Carlo estimation of exact p-values. You can request Monte Carlo estimation by specifying the MC computation-option in the EXACT statement. See the section Computational Resources for more information.

FORMAT=format-name

specifies a format for the following crosstabulation table cell values: frequency, expected frequency, and deviation. PROC FREQ also uses the specified format to display the row and column total frequencies and the overall total frequency in crosstabulation tables.

You can specify any standard SAS numeric format or a numeric format defined with the FORMAT procedure. The format length must not exceed 24. If you omit the FORMAT= option, by default PROC FREQ uses the BEST6. format to display frequencies less than 1E6, and the BEST7. format otherwise.

The FORMAT= option applies only to crosstabulation tables displayed in the default format. It does not apply to crosstabulation tables produced with the LIST or CROSSLIST option.

To change display formats in any FREQ table, you can use PROC TEMPLATE. See the chapter “The TEMPLATE Procedure” in the SAS Output Delivery System: User's Guide for more information.

GAILSIMON < (COLUMN=1 | 2)> GS < (COLUMN=1 | 2)>

requests the Gail-Simon test for qualitative interaction, which applies to stratified $2 \times 2$ tables. See the section Gail-Simon Test for Qualitative Interactions for details.

JT

requests the Jonckheere-Terpstra test. See the section Jonckheere-Terpstra Test for details. To request exact p-values for the Jonckheere-Terpstra test, specify the JT option in the EXACT statement. See the section Exact Statistics for more information.

LIST

displays two-way and multiway tables by using a list format instead of the default crosstabulation cell format. This option displays an entire multiway table in one table, instead of displaying a separate two-way table for each stratum. For more information, see the section Two-Way and Multiway Tables.

The LIST option is not available when you request tests and statistics; you must use the standard crosstabulation table display or the CROSSLIST display when you request tests and statistics.

MAXLEVELS=n

specifies the maximum number of variable levels to display in one-way frequency tables. The value of n must be a positive integer. PROC FREQ displays the first n variable levels, matching the order in which the levels appear in the one-way frequency table. (The ORDER= option controls the order of the variable levels. By default, ORDER=INTERNAL, which orders the variable levels by unformatted value.)

The MAXLEVELS= option also applies to one-way frequency plots, which you can request by specifying the PLOTS=FREQPLOT option when ODS Graphics is enabled.

If you specify the MISSPRINT option to display missing levels in the frequency table, the MAXLEVELS= option displays the first n nonmissing levels.

The MAXLEVELS= option does not apply to the OUT= output data set, which includes all variable levels. The MAXLEVELS= option does not affect the computation of percentages, statistics, or tests for the one-way table; these values are based on the complete table.

MEASURES < (CL)>

requests measures of association and their asymptotic standard errors. This option provides the following measures: gamma, Kendall’s tau-b, Stuart’s tau-c, Somers’ $D (C|R)$ , Somers’ $D (R|C)$ , Pearson and Spearman correlation coefficients, lambda (symmetric and asymmetric), and uncertainty coefficients (symmetric and asymmetric). If you specify the CL option in parentheses after the MEASURES option, PROC FREQ provides confidence limits for the measures of association. For more information, see the section Measures of Association.

For $2 \times 2$ tables, the MEASURES option also provides the odds ratio, column 1 relative risk, column 2 relative risk, and their asymptotic Wald confidence limits. You can request the odds ratio and relative risks separately (without the other measures of association) by specifying the RELRISK option. You can request confidence limits for the odds ratio by specifying the OR(CL=) option.

You can use the TEST statement to request asymptotic tests for the following measures of association: gamma, Kendall’s tau-b, Stuart’s tau-c, Somers’ $D (C|R)$ , Somers’ $D (R|C)$ , and Pearson and Spearman correlation coefficients. You can use the EXACT statement to request exact confidence limits for the odds ratio, exact unconditional confidence limits for the relative risks, and exact tests for the following measures of association: Kendall’s tau-b, Stuart’s tau-c, Somers’ $D (C|R)$ and $D (R|C)$ , and Pearson and Spearman correlation coefficients. For more information, see the descriptions of the TEST and EXACT statements and the section Exact Statistics.

MISSING

treats missing values as a valid nonmissing level for all TABLES variables. The MISSING option displays the missing levels in frequency and crosstabulation tables and includes them in all calculations of percentages, tests, and measures.

By default, if you do not specify the MISSING or MISSPRINT option, an observation is excluded from a table if it has a missing value for any of the variables in the TABLES request. When PROC FREQ excludes observations with missing values, it displays the total frequency of missing observations below the table. See the section Missing Values for more information.

MISSPRINT

displays missing value frequencies in frequency and crosstabulation tables but does not include the missing value frequencies in any computations of percentages, tests, or measures.

NOCOL

suppresses the display of column percentages in crosstabulation table cells.

NOCUM

suppresses the display of cumulative frequencies and percentages in one-way frequency tables. The NOCUM option also suppresses the display of cumulative frequencies and percentages in crosstabulation tables in list format, which you request with the LIST option.

NOFREQ

suppresses the display of cell frequencies in crosstabulation tables. The NOFREQ option also suppresses row total frequencies. This option has no effect for one-way tables or for crosstabulation tables in list format, which you request with the LIST option.

NOPERCENT

suppresses the display of overall percentages in crosstabulation tables. These percentages include the cell percentages of the total (two-way) table frequency, as well as the row and column percentages of the total table frequency. To suppress the display of cell percentages of row or column totals, use the NOROW or NOCOL option, respectively.

For one-way frequency tables and crosstabulation tables in list format, the NOPERCENT option suppresses the display of percentages and cumulative percentages.

NOPRINT

suppresses the display of frequency and crosstabulation tables but displays all requested tests and statistics. To suppress the display of all output, including tests and statistics, use the NOPRINT option in the PROC FREQ statement.

NOROW

suppresses the display of row percentages in crosstabulation table cells.

NOSPARSE

suppresses the display of cells with a zero frequency count in LIST output and omits them from the OUT= data set. The NOSPARSE option applies when you specify the ZEROS option in the WEIGHT statement to include observations with zero weights. By default, the ZEROS option invokes the SPARSE option, which displays table cells with a zero frequency count in the LIST output and includes them in the OUT= data set. See the description of the ZEROS option for more information.

The NOSPARSE option also suppresses the display of variable levels with zero frequency in CROSSLIST tables. By default for CROSSLIST tables, PROC FREQ displays all levels of the column variable within each level of the row variable, including any column variable levels with zero frequency for that row. For multiway tables displayed with the CROSSLIST option, the procedure displays all levels of the row variable for each stratum of the table by default, including any row variable levels with zero frequency for the stratum.

NOWARN

suppresses the log warning message for the validity of the asymptotic Pearson chi-square test. By default, PROC FREQ provides a validity warning for the asymptotic Pearson chi-square test when more than 20cells have expected frequencies that are less than 5. This warning message appears in the log if you specify the NOPRINT option in the PROC FREQ statement,

The NOWARN option is equivalent to the CHISQ(WARN=NOLOG) option. You can also use the CHISQ(WARN=) option to suppress the warning message in the display and to request a warning variable in the chi-square ODS output data set or in the OUTPUT data set.

OR(CL=type | (types)) ODDSRATIO(CL=type | (types))

requests the odds ratio and confidence limits for $2 \times 2$ tables. You can specify one or more types of confidence limits, which include exact, score, and Wald confidence limits. When you specify only one confidence limit type, you can omit the parentheses around the request.

PROC FREQ displays the confidence limits in the “Confidence Limits for the Odds Ratio” table. Specifying the OR option without the CL= option is equivalent to specifying the RELRISK option, which produces the “Odds Ratio and Relative Risks” table. For more information, see the description of the RELRISK option. When you specify the OR(CL=) option, PROC FREQ does not produce the “Odds Ratio and Relative Risks” table unless you also specify the RELRISK or MEASURES option.

The ALPHA= option determines the confidence level, and the default of ALPHA=0.05 produces 95% confidence limits for the odds ratio.

You can specify the following types:

EXACT: displays exact confidence limits for the odds ratio in the “Confidence Limits for the Odds Ratio” table. You must also request computation of the exact confidence limits by specifying the OR option in the EXACT statement. For more information, see the section Exact Confidence Limits for the Odds Ratio.
SCORE < (CORRECT=NO)>: requests score confidence limits for the odds ratio. For more information, see the section Score Confidence Limits for the Odds Ratio. If you specify CORRECT=NO, PROC FREQ provides the uncorrected form of the confidence limits.
WALD: requests asymptotic Wald confidence limits for the odds ratio. For more information, see the section Odds Ratio and Relative Risks for 2 x 2 Tables.

OUT=SAS-data-set

names an output data set that contains frequency or crosstabulation table counts and percentages. If more than one table request appears in the TABLES statement, the contents of the OUT= data set correspond to the last table request in the TABLES statement. The OUT= data set variable COUNT contains the frequencies and the variable PERCENT contains the percentages. See the section Output Data Sets for details. You can specify the following options to include additional information in the OUT= data set: OUTCUM, OUTEXPECT, and OUTPCT.

OUTCUM

includes cumulative frequencies and cumulative percentages in the OUT= data set for one-way tables. The variable CUM_FREQ contains the cumulative frequencies, and the variable CUM_PCT contains the cumulative percentages. See the section Output Data Sets for details. The OUTCUM option has no effect for two-way or multiway tables.

OUTEXPECT

includes expected cell frequencies in the OUT= data set for crosstabulation tables. The variable EXPECTED contains the expected cell frequencies. See the section Output Data Sets for details. The EXPECTED option has no effect for one-way tables.

OUTPCT

includes the following additional variables in the OUT= data set for crosstabulation tables:

PCT_COL: percentage of column frequency
PCT_ROW: percentage of row frequency
PCT_TABL: percentage of stratum (two-way table) frequency, for n-way tables where n > 2

See the section Output Data Sets for details. The OUTPCT option has no effect for one-way tables.

PLCORR < (options)>

requests the polychoric correlation coefficient and its asymptotic standard error. For $2 \times 2$ tables, this statistic is more commonly known as the tetrachoric correlation coefficient, and it is labeled as such in the displayed output. For more information, see the section Polychoric Correlation.

If you also specify the CL or MEASURES(CL) option, PROC FREQ provides confidence limits for the polychoric correlation. If you specify the PLCORR option in the TEST statement, the procedure provides Wald and likelihood ratio tests for the polychoric correlation. The PLCORR option invokes the MEASURES option.

You can specify the following options:

CONVERGE=value: specifies the convergence criterion for computing the polychoric correlation. The convergence criterion value must be a positive number. By default, CONVERGE=0.0001. Iterative computation of the polychoric correlation stops when the convergence measure falls below value or when the number of iterations exceeds the MAXITER= number, whichever happens first. For parameter values that are less than 0.01, PROC FREQ evaluates convergence by using the absolute difference instead of the relative difference. For more information, see the section Polychoric Correlation.
MAXITER=number: specifies the maximum number of iterations for computing the polychoric correlation. The value of number must be a positive integer. By default, MAXITER=20. Iterative computation of the polychoric correlation stops when the number of iterations exceeds the maximum number or when the convergence measure falls below the CONVERGE= value, whichever happens first. For more information, see the section Polychoric Correlation.

PLOTS < (global-plot-options)> < =plot-request < (plot-options)> > PLOTS < (global-plot-options)> < =(plot-request < (plot-options)> < …plot-request < (plot-options)> > )>

controls the plots that are produced through ODS Graphics. Plot-requests identify the plots, and plot-options control the appearance and content of the plots. You can specify plot-options in parentheses after a plot-request. A global-plot-option applies to all plots for which it is available unless it is altered by a specific plot-option. You can specify global-plot-options in parentheses after the PLOTS option.

When you specify only one plot-request, you can omit the parentheses around the request. For example:

 plots=all
 plots=freqplot
 plots=(freqplot oddsratioplot)
 plots(only)=(cumfreqplot deviationplot)

ODS Graphics must be enabled before plots can be requested. For example:

ods graphics on;
proc freq;
   tables treatment*response / chisq plots=freqplot;
   weight wt;
run;
ods graphics off;

For more information about enabling and disabling ODS Graphics, see the section “Enabling and Disabling ODS Graphics” in Chapter 21: Statistical Graphics Using ODS in SAS/STAT User's Guide.

If ODS Graphics is enabled but you do not specify the PLOTS= option, PROC FREQ produces all plots that are associated with the analyses that you request, with the exception of the frequency, cumulative frequency, and mosaic plots. To produce a frequency plot or cumulative frequency plot when ODS Graphics is enabled, you must specify the FREQPLOT or CUMFREQPLOT plot-request, respectively, in the PLOTS= option, or you must specify the PLOTS=ALL option. To produce a mosaic plot when ODS Graphics is enabled, you must specify the MOSAICPLOT plot-request in the PLOTS= option, or you must specify the PLOTS=ALL option.

PROC FREQ produces the remaining plots (listed in Table 3.11) by default when you request the corresponding TABLES statement options. You can suppress default plots and request specific plots by using the PLOTS(ONLY)= option; PLOTS(ONLY)=(plot-requests) produces only the plots that are specified as plot-requests. You can suppress all plots by specifying the PLOTS=NONE option. The PLOTS option has no effect when you specify the NOPRINT option in the PROC FREQ statement.

Plot Requests

Table 3.11 lists the available plot-requests together with their required TABLES statement options. Descriptions of the plot-requests follow the table in alphabetical order.

Table 3.11: Plot Requests

Plot Request	Description	Required TABLES Statement Option
AGREEPLOT	Agreement plot	AGREE ( $r \times r$ table)
ALL	All plots	None
CUMFREQPLOT	Cumulative frequency plot	One-way table request
DEVIATIONPLOT	Deviation plot	CHISQ (one-way table)
FREQPLOT	Frequency plot	Any table request
KAPPAPLOT	Kappa plot	AGREE ( $h \times r \times r$ table)
MOSAICPLOT	Mosaic plot	Two-way or multiway table request
NONE	No plots	None
ODDSRATIOPLOT	Odds ratio plot	MEASURES or RELRISK ( $h \times 2 \times 2$ table)
RELRISKPLOT	Relative risk plot	MEASURES or RELRISK ( $h \times 2 \times 2$ table)
RISKDIFFPLOT	Risk difference plot	RISKDIFF ( $h \times 2 \times 2$ table)
WTKAPPAPLOT	Weighted kappa plot	AGREE ( $h \times r \times r$ table, $r > 2$ )

The following plot-requests are available with the PLOTS= option:

AGREEPLOT < (plot-options)>

requests an agreement plot (Bangdiwala and Bryan, 1987), An agreement plot displays the strength of agreement in a two-way table, where the row and column variables represent two independent ratings of n subjects. For information about agreement plots, see Bangdiwala (1988); Bangdiwala et al. (2008), and Friendly (2000, Section 3.7.2).

To produce an agreement plot, you must also specify the AGREE option in the TABLES statement. Agreement statistics and plots are available for two-way square tables, where the number of rows equals the number of columns.

Table 3.12 lists the plot-options that are available for agreement plots. See the subsection “Plot Options” for descriptions of the plot-options.

Table 3.12: Plot Options for AGREEPLOT

Plot Option	Description	Values
LEGEND=	Legend	YES $^*$ or NO
PARTIAL=	Partial agreement	YES $^*$ or NO
SHOWSCALE=	Frequency scale	YES $^*$ or NO
STATS	Statistics	None
$^*$ Default

ALL

requests all plots that are associated with the specified analyses. Table 3.11 lists the available plot-requests and the corresponding analysis options. If you specify the PLOTS=ALL option, PROC FREQ produces the frequency, cumulative frequency, and mosaic plots that are associated with the tables that you request. (These plots are not produced by default when ODS Graphics is enabled.)

CUMFREQPLOT < (plot-options)>

requests a plot of cumulative frequencies. Cumulative frequency plots are available for one-way frequency tables.

To produce a cumulative frequency plot, you must specify the CUMFREQPLOT plot-request in the PLOTS= option, or you must specify the PLOTS=ALL option. PROC FREQ does not produce cumulative frequency plots by default when ODS Graphics is enabled.

Table 3.13 lists the plot-options that are available for cumulative frequency plots. See the subsection “Plot Options” for descriptions of the plot-options.

Table 3.13: Plot Options for CUMFREQPLOT

Plot Option	Description	Values
ORIENT=	Orientation	VERTICAL $^*$ or HORIZONTAL
SCALE=	Scale	FREQ $^*$ or PERCENT
TYPE=	Type	BARCHART $^*$ or DOTPLOT
$^*$ Default

DEVIATIONPLOT < (plot-options)>

requests a plot of relative deviations from expected frequencies. Deviation plots are available for chi-square analysis of one-way frequency tables. To produce a deviation plot, you must also specify the CHISQ option in the TABLES statement for a one-way frequency table.

Table 3.14 lists the plot-options that are available for deviation plots. See the subsection “Plot Options” for descriptions of the plot-options.

Table 3.14: Plot Options for DEVIATIONPLOT

Plot Option	Description	Values
NOSTAT	No statistic	None
ORIENT=	Orientation	VERTICAL $^*$ or HORIZONTAL
TYPE=	Type	BARCHART $^*$ or DOTPLOT
$^*$ Default

FREQPLOT < (plot-options)>

requests a frequency plot. Frequency plots are available for frequency and crosstabulation tables. For multiway crosstabulation tables, PROC FREQ provides a two-way frequency plot for each stratum (two-way table).

To produce a frequency plot, you must specify the FREQPLOT plot-request in the PLOTS= option, or you must specify the PLOTS=ALL option. PROC FREQ does not produce frequency plots by default when ODS Graphics is enabled.

By default, PROC FREQ displays frequency plots as bar charts. You can specify the TYPE=DOTPLOT plot-option to display frequency plots as dot plots. You can plot percentages instead of frequencies by specifying the SCALE=PERCENT plot-option. There are four frequency plot layouts available, which you can request by specifying the TWOWAY= plot-option. See the subsection “Plot Options” for more information.

By default, the primary grouping of graph cells in a two-way layout is by column variable. Row variable levels are then displayed within column variable levels. You can specify the GROUPBY=ROW plot-option to group first by row variable.

Table 3.15 lists the plot-options that are available for frequency plots. See the subsection “Plot Options” for descriptions of the plot-options.

The following plot-options are available for all frequency plots: ORIENT=, SCALE=, and TYPE=. The following plot-options are available for frequency plots of two-way (and multiway) tables: GROUPBY=, NPANELPOS=, and TWOWAY=. The NPANELPOS= plot-option is not available with the TWOWAY=CLUSTER or TWOWAY=STACKED layout, which is always displayed in a single panel.

Table 3.15: Plot Options for FREQPLOT

Plot Option	Description	Values
GROUPBY= $^{**}$	Primary group	COLUMN $^*$ or ROW
NPANELPOS= $^{**}$	Sections per panel	Number (4 $^*$ )
ORIENT=	Orientation	VERTICAL $^*$ or HORIZONTAL
SCALE=	Scale	FREQ $^$ , GROUPPERCENT $^{*}$ ,
		LOG, PERCENT, SQRT
TWOWAY= $^{**}$	Two-way layout	GROUPVERTICAL $^*$ CLUSTER,
		GROUPHORIZONTAL, or STACKED
TYPE=	Type	BARCHART $^*$ or DOTPLOT
$^*$ Default
$^{**}$ For two-way tables

KAPPAPLOT < (plot-options)>

requests a plot of kappa statistics with confidence limits. Kappa plots are available for multiway square tables and display the kappa statistic (with confidence limits) for each two-way table (stratum). Kappa plots also display the overall kappa statistic unless you specify the COMMON=NO plot-option. To produce a kappa plot, you must specify the AGREE option in the TABLES statement to compute kappa statistics.

Table 3.16 lists the plot-options that are available for kappa plots. See the subsection “Plot Options” for descriptions of the plot-options.

Table 3.16: Plot Options for KAPPAPLOT and WTKAPPAPLOT

Plot Option	Description	Values
CLDISPLAY=	Error bar type	SERIF $^*$ , SERIFARROW,
		LINE, LINEARROW, or BAR
COMMON=	Overall kappa	YES $^*$ or NO
NPANELPOS=	Statistics per graphic	Number (All $^*$ )
ORDER=	Order of two-way levels	ASCENDING or DESCENDING
RANGE=	Range to display	Values or CLIP
STATS	Statistic values	None
$^*$ Default

MOSAICPLOT < (plot-options)>

requests a mosaic plot. Mosaic plots are available for two-way and multiway crosstabulation tables; for multiway tables, PROC FREQ provides a mosaic plot for each two-way table (stratum).

To produce a mosaic plot, you must specify the MOSAICPLOT plot-request in the PLOTS= option, or you must specify the PLOTS=ALL option. PROC FREQ does not produce mosaic plots by default when ODS Graphics is enabled.

Mosaic plots display tiles that correspond to the crosstabulation table cells. The areas of the tiles are proportional to the frequencies of the table cells. The column variable is displayed on the X axis, and the tile widths are proportional to the relative frequencies of the column variable levels. The row variable is displayed on the Y axis, and the tile heights are proportional to the relative frequencies of the row levels within column levels. For more information, see Friendly (2000).

By default, the colors of the tiles correspond to the row variable levels. If you specify the COLORSTAT= plot-option, the tiles are colored according to the values of the Pearson or standardized residuals.

You can specify the following plot-options:

COLORSTAT < =PEARSONRES | STDRES>: colors the mosaic plot tiles according to the values of residuals. If you specify COLORSTAT=PEARSONRES, the tiles are colored according to the Pearson residuals of the corresponding table cells. For more information, see the section Pearson Chi-Square Test for Two-Way Tables. If you specify COLORSTAT=STDRES, the tiles are colored according to the standardized residuals of the corresponding table cells. For more information, see the section Standardized Residuals. You can display the Pearson or standardized residuals in the CROSSLIST table by specifying the CROSSLIST(PEARSONRES) or CROSSLIST(STDRES) option, respectively.
SQUARE: produces a square mosaic plot, where the height of the Y axis equals the width of the X axis. In a square mosaic plot, the scale of the relative frequencies is the same on both axes. By default, PROC FREQ produces a rectangular mosaic plot.

NONE

suppresses all plots.

ODDSRATIOPLOT < (plot-options)>

requests a plot of odds ratios with confidence limits. Odds ratio plots are available for multiway $2 \times 2$ tables and display the odds ratio (with confidence limits) for each $2 \times 2$ table (stratum). To produce an odds ratio plot, you must also specify the MEASURES, OR(CL=), or RELRISK option in the TABLES statement to compute odds ratios.

By default, the odds ratio plot displays asymptotic Wald confidence limits for the odds ratios (CL=WALD). You can specify the CL=EXACT or CL=SCORE plot-option to display exact or score confidence limits, respectively. To display exact or score confidence limits, you must also request their computation. You can specify the OR option in the EXACT statement to request exact confidence limits for the odds ratio; you can specify the OR(CL=SCORE) option in the TABLES statement to request score confidence limits for the odds ratio.

By default, the odds ratio plot displays the common odds ratio when the procedure computes the common value. You can suppress display of the common odds ratio by specifying the COMMON=NO plot-option. To compute the common odds ratio for a multiway table, you can specify the CMH option in the TABLES statement. To display the common odds ratio when the plot displays exact confidence limits (CL=EXACT), you must request computation of exact confidence limits for the common odds ratio by specifying the COMOR option in the EXACT statement. When you specify the CL=SCORE plot-option, the odds ratio plot does not display the common odds ratio.

Table 3.17 lists the plot-options that are available for odds ratio plots. See the subsection “Plot Options” for descriptions of the plot-options.

Table 3.17: Plot Options for ODDSRATIOPLOT, RELRISKPLOT, and RISKDIFFPLOT

Plot Option	Description	Values
CL=	Confidence limit type	Type
CLDISPLAY=	Error bar type	SERIF $^*$ , SERIFARROW,
		LINE, LINEARROW, or BAR
COMMON=	Common value	YES $^*$ or NO
COLUMN= $^{**}$	Risk column	1 $^*$ or 2
LOGBASE= $^{***}$	Axis scale	2, E, or 10
NPANELPOS=	Statistics per graphic	Number (All $^*$ )
ORDER=	Order of two-way levels	ASCENDING or DESCENDING
RANGE=	Range to display	Values or CLIP
STATS	Statistic values	None
$^*$ Default
$^{**}$ Available for RELRISKPLOT and RISKDIFFPLOT
$^{***}$ Available for ODDSRATIOPLOT and RELRISKPLOT

RELRISKPLOT < (plot-options)>

requests a plot of relative risks with confidence limits. Relative risk plots are available for multiway $2 \times 2$ tables and display the relative risk (with confidence limits) for each $2 \times 2$ table (stratum). To produce a relative risk plot, you must also specify the MEASURES or RELRISK option in the TABLES statement to compute relative risks.

If you specify the CMH option in the TABLES statement to compute the common relative risk for the multiway table, by default the plot displays the common relative risk. The COMMON=NO plot-option suppresses display of the common relative risk. If you do not specify the CMH option, the common relative risk is not available to be displayed in the plot.

By default, relative risk plots display asymptotic confidence limits. You can specify the CL=EXACT plot-option to display exact confidence limits for the relative risks. You must also request computation of exact confidence limits by specifying the RELRISK option in the EXACT statement. When you specify the CL=EXACT plot-option, the relative risk plot does not display the common relative risk.

Table 3.17 lists the plot-options that are available for relative risk plots. See the subsection “Plot Options” for descriptions of the plot-options.

RISKDIFFPLOT < (plot-options)>

requests a plot of risk (proportion) differences with confidence limits. Risk difference plots are available for multiway $2 \times 2$ tables and display the risk difference (with confidence limits) for each $2 \times 2$ table (stratum). To produce a risk difference plot, you must also specify the RISKDIFF option in the TABLES statement to compute risk differences.

By default, risk difference plots display asymptotic Wald confidence limits. You can specify the CL=EXACT plot-option to display exact confidence limits for the risk differences. You must also request computation of exact confidence limits by specifying the RISKDIFF option in the EXACT statement.

In addition to Wald and exact confidence limits, you can display the following confidence limit types in a risk difference plot: Agresti-Caffo, Hauck-Anderson, Miettinen-Nurminen (score), and Newcombe. See the descriptions of the CL= plot-option and the RISKDIFF option for more information.

By default or when you specify CL=WALD, the risk difference plot displays the common risk difference when PROC FREQ computes the common value. To compute the common risk difference for a multiway table, you can specify the RISKDIFF(COMMON) option in the TABLES statement. When you specify other confidence limit types, the risk difference plot does not display the common risk difference. You can suppress display of the common risk difference by specifying the COMMON=NO plot-option.

Table 3.17 lists the plot-options that are available for risk difference plots. See the subsection “Plot Options” for descriptions of the plot-options.

WTKAPPAPLOT < (plot-options)>

requests a plot of weighted kappa coefficients with confidence limits. Weighted kappa plots are available for multiway square tables and display the weighted kappa coefficient (with confidence limits) for each two-way table (stratum). Weighted kappa plots also display the overall weighted kappa coefficient unless you specify the COMMON=NO plot-option.

To produce a weighted kappa plot, you must specify the AGREE option in the TABLES statement to compute weighted kappa coefficients, and the table dimension must be greater than 1.

Table 3.16 lists the plot-options that are available for weighted kappa plots. See the subsection “Plot Options” for descriptions of the plot-options.

Global Plot Options

A global-plot-option applies to all plots for which the option is available unless it is altered by an individual plot-option. You can specify global-plot-options in parentheses after the PLOTS option. For example:

 plots(order=ascending stats)=(riskdiffplot oddsratioplot)
 plots(only)=freqplot

The following plot-options are available as global-plot-options: CLDISPLAY=, COLUMN=, COMMON=, EXACT, LOGBASE=, NPANELPOS=, ORDER=, ORIENT=, RANGE=, SCALE=, STATS, and TYPE=. See the subsection “Plot Options” for description of these plot-options.

In addition to these plot-options, you can specify the following global-plot-option:

ONLY: suppresses the default plots and requests only the plots that are specified as plot-requests.

Plot Options

You can specify the following plot-options in parentheses after a plot-request:

CL=type

specifies the type of confidence limits to display. The CL= plot-option is available for the following plots: ODDSRATIOPLOT, RELRISKPLOT, and RISKDIFFPLOT.

For odds ratio plots, the available confidence limit types include exact, score, and Wald, which you can request by specifying CL=EXACT, CL=SCORE, and CL=WALD, respectively. The default confidence limit type for odds ratio plots is CL=WALD. When you specify CL=EXACT to display exact confidence limits, you must also request computation of exact confidence limits by specifying the OR option in the EXACT statement. When you specify CL=SCORE, you must request computation of score confidence limits by specifying the OR(CL=SCORE) option in the TABLES statement.

For relative risk plots, the available confidence limit types include exact and Wald, which you can request by specifying CL=EXACT and CL=WALD, respectively. The default is CL=WALD. When you specify CL=EXACT to display exact confidence limits, you must also request computation of exact confidence limits by specifying the RELRISK option in the EXACT statement.

For risk difference plots, the available confidence limit types include the following: CL=AC (Agresti-Caffo), CL=EXACT (exact unconditional), CL=HA (Hauck-Anderson), CL=MN (Miettinen-Nurminen (score)), CL=NEWCOMBE (Newcombe), and CL=WALD (Wald). For more information, see the description of the CL= riskdiff-option and the section Risk Difference Confidence Limits. If you specify CL=EXACT to display exact confidence limits in the plot, you must also request computation of exact confidence limits by specifying the RISKDIFF option in the EXACT statement.

CLDISPLAY=SERIF | SERIFARROW | LINE | LINEARROW | BAR < width >

controls the appearance of the confidence limit error bars. This plot-option is available for the following plots: KAPPAPLOT, ODDSRATIOPLOT, RELRISKPLOT, RISKDIFFPLOT, and WTKAPPAPLOT.

The default is CLDISPLAY=SERIF, which displays the confidence limits as lines with serifs. CLDISPLAY=LINE displays the confidence limits as plain lines without serifs. The CLDISPLAY=SERIFARROW and CLDISPLAY=LINEARROW plot-options display arrowheads on any error bars that are clipped by the RANGE= plot-option; if an entire error bar is cut from the plot, the plot displays an arrowhead that points toward the statistic.

CLDISPLAY=BAR displays the confidence limits as bars. By default, the width of the bars equals the size of the marker for the estimate. You can control the width of the bars and the size of the marker by specifying the value of width as a percentage of the distance between bars, $0 < \Argument{width} \le 1$ . The bar might disappear when the value of width is very small.

COLUMN=1 | 2

specifies the table column to use to compute the risks (proportion) for the relative risk plot (RELRISKPLOT) and the risk difference plot (RISKDIFFPLOT). If you specify COLUMN=1, the plot displays the column 1 relative risks or the column 1 risk differences. Similarly, if you specify COLUMN=2, the plot displays the column 2 relative risks or risk differences.

For relative risk plots, the default is COLUMN=1. For risk difference plots, the default is COLUMN=1 if you request computation of both column 1 and column 2 risk differences by specifying the RISKDIFF option. If you request computation of only the column 1 (or column 2) risk differences by specifying the RISKDIFF(COLUMN=1) (or RISKDIFF(COLUMN=2)) option, by default the risk difference plot displays the risk differences for the column that you specify.

COMMON=YES | NO

controls the display of the common (overall) statistic in plots that display stratum (two-way table) statistics for a multiway table. The COMMON= plot-option is available for the following plots: KAPPAPLOT, ODDSRATIOPLOT, RELRISKPLOT, RISKDIFFPLOT, and WTKAPPAPLOT.

By default, COMMON=YES and plots display the common value when PROC FREQ computes it. When you specify the AGREE option in the TABLES statement for a multiway table, PROC FREQ computes the common simple and weighted kappa coefficients, which can be displayed in the simple and weighted kappa plots, respectively. When you specify the MEASURES or RELRISK option in the TABLES statement, PROC FREQ computes only stratum-level odds ratios and relative risks; to compute common odds ratios and relative risks, you must also specify the CMH option in the TABLES statement. To compute the common risk difference, you must specify the RISKDIFF(COMMON) option. For more information, see the descriptions of the plot-requests.

EXACT

requests display of exact confidence limits instead of asymptotic confidence limits. The EXACT plot-option is available for the odds ratio plot (ODDSRATIOPLOT), relative risk plot (RELRISKPLOT), and risk difference plot (RISKDIFFPLOT). The EXACT plot-option is equivalent to the CL=EXACT plot-option.

When you specify the EXACT plot-option, you must also request computation of exact confidence limits by specifying the appropriate statistic-option in the EXACT statement.

GROUPBY=COLUMN | ROW

specifies the primary grouping for two-way frequency plots, which you can request by specifying the FREQPLOT plot=request. The default is GROUPBY=COLUMN, which groups graph cells first by column variable and displays row variable levels within column variable levels. You can specify GROUPBY=ROW to group first by row variable. In two-way and multiway table requests, the column variable is the last variable specified and forms the columns of the crosstabulation table. The row variable is the next-to-last variable specified and forms the rows of the table.

By default for a bar chart that is displayed in the TWOWAY=STACKED layout, bars correspond to the column variable levels, and row levels are displayed (stacked) within each column bar. By default for a bar chart that is displayed in the TWOWAY=CLUSTER layout, bars are first grouped by column variable levels, and row levels are displayed as adjacent bars within each column-level group. You can reverse the default row and column variable grouping by specifying GROUPBY=ROW.

LOGBASE=2 | E | 10

applies to the odds ratio plot (ODDSRATIOPLOT) and the relative risk plot (RELRISKPLOT). This plot-option displays the odds ratio or relative risk axis on the log scale that you specify.

LEGEND=YES | NO

applies to the agreement plot (AGREEPLOT). LEGEND=NO suppresses the legend that identifies the areas of exact and partial agreement. The default is LEGEND=YES.

NOSTAT

applies to the deviation plot (DEVIATIONPLOT). NOSTAT suppresses the chi-square p-value that deviation plot displays by default.

NPANELPOS=n

divides the plot into multiple panels that display at most $|\Argument{n}|$ statistics or sections.

If n is positive, the number of statistics or sections per panel is balanced; if n is negative, the number of statistics per panel is not balanced. For example, suppose you want to display 21 odds ratios. NPANELPOS=20 displays two panels, the first with 11 odds ratios and the second with 10 odds ratios; NPANELPOS=–20 displays 20 odds ratios in the first panel but only 1 odds ratio in the second panel. This plot-option is available for all plots except mosaic plots and one-way weighted frequency plots.

For two-way frequency plots (FREQPLOT), NPANELPOS=n requests that panels display at most $|\Argument{n}|$ sections, where sections correspond to row or column variable levels, depending on the type of plot and the grouping. By default, n=4 and each panel includes at most four sections. This plot-option applies to two-way plots that are displayed in the TWOWAY=GROUPVERTICAL or TWOWAY=GROUPHORIZONTAL layout. The NPANELPOS= plot-option does not apply to the TWOWAY=CLUSTER and TWOWAY=STACKED layouts, which are always displayed in a single panel.

For plots that display statistics along with confidence limits, NPANELPOS=n requests that panels display at most $|\Argument{n}|$ statistics. By default, n=0 and all statistics are displayed in a single panel. This plot-option applies to the following plots: KAPPAPLOT, ORPLOT, RELRISKPLOT, RISKDIFFPLOT, and WTKAPPAPLOT.

ORDER=ASCENDING | DESCENDING

displays the two-way table (strata) statistics in order of the statistic value. If you specify ORDER=ASCENDING or ORDER=DESCENDING, the plot displays the statistics in ascending or descending order, respectively. By default, the order of the statistics in the plot matches the order that the two-way table strata appear in the multiway table display.

This plot-option applies to the following plots: KAPPAPLOT, ODDSRATIOPLOT, RELRISKPLOT, RISKDIFFPLOT, and WTKAPPAPLOT.

ORIENT=HORIZONTAL | VERTICAL

controls the orientation of the plot. The ORIENT= plot-option applies to the following plots: CUMFREQPLOT, DEVIATIONPLOT, and FREQPLOT.

ORIENT=HORIZONTAL places the variable levels on the Y axis and the frequencies, percentages, or statistic values on the X axis. ORIENT=VERTICAL places the variable levels on the X axis. The default orientation is ORIENT=VERTICAL for bar charts (TYPE=BARCHART) and ORIENT=HORIZONTAL for dot plots (TYPE=DOTPLOT).

PARTIAL=YES | NO

controls the display of partial agreement in the agreement plot (AGREEPLOT). PARTIAL=NO suppresses the display of partial agreement. When you specify PARTIAL=NO, the agreement plot displays only exact agreement. Exact agreement includes the diagonal cells of the square table, where the row and column variable levels are the same. Partial agreement includes the adjacent off-diagonal table cells, where the row and column values are within one level of exact agreement. The default is PARTIAL=YES.

RANGE=( < min > < , max > )| CLIP

specifies the range of values to display. If you specify RANGE=CLIP, the confidence limits are clipped and the display range is determined by the minimum and maximum values of the statistics. By default, the display range includes all confidence limits.

This plot-option is available for the following plots: KAPPAPLOT, ODDSRATIOPLOT, RELRISKPLOT, RISKDIFFPLOT, and WTKAPPAPLOT.

SCALE=FREQ | GROUPPERCENT | LOG | PERCENT | SQRT

specifies the scale of the frequencies to display. This plot-option is available for frequency plots (FREQPLOT) and cumulative frequency plots (CUMFREQPLOT).

The default is SCALE=FREQ, which displays unscaled frequencies. SCALE=PERCENT displays percentages (relative frequencies) of the total frequency. SCALE=LOG displays log (base 10) frequencies. SCALE=SQRT displays square roots of the frequencies, producing a plot known as a rootogram.

SCALE=GROUPPERCENT is available for two-way frequency plots. This option displays the row or column percentages instead of the overall percentages (of the table frequency). By default (or when you specify the GROUPBY=COLUMN plot-option), SCALE=GROUPPERCENT displays the column percentages. If you specify the GROUPBY=ROW plot-option, the primary grouping of graph cells is by row variable level and the plot displays row percentages. For more information, see the description of the GROUPBY= plot-option.

SHOWSCALE=YES | NO

controls the display of the cumulative frequency scale on the right side of the agreement plot (AGREEPLOT). SHOWSCALE=NO suppresses the display of the scale. The default is SHOWSCALE=YES.

STATS

displays statistic values in the plot. For the following plots, the STATS plot-option displays the statistics and their confidence limits on the right side of the plot: KAPPAPLOT, ODDSRATIOPLOT, RELRISKPLOT, RISKDIFFPLOT, and WTKAPPAPLOT.

For the agreement plot (AGREEPLOT), STATS displays the values of the kappa statistic, the weighted kappa statistic, and the $B_ n$ measure (Bangdiwala and Bryan, 1987).

If you do not request the STATS plot-option, these plots do not display the statistic values.

TWOWAY=CLUSTER | GROUPHORIZONTAL | GROUPVERTICAL | STACKED

specifies the layout for two-way frequency plots.

All TWOWAY= layouts are available for bar charts (TYPE=BARCHART). All TWOWAY= layouts except TWOWAY=CLUSTER are available for dot plots (TYPE=DOTPLOT). The ORIENT= and GROUPBY= plot-options are available for all TWOWAY= layouts.

The default two-way layout is TWOWAY=GROUPVERTICAL, which produces a grouped plot that has a vertical common baseline. By default for bar charts (TYPE=BARCHART, ORIENT=VERTICAL), the X axis displays column variable levels, and the Y axis displays frequencies. The plot includes a vertical (Y-axis) block for each row variable level. The relative positions of the graph cells in this plot layout are the same as the relative positions of the table cells in the crosstabulation table. You can reverse the default row and column grouping by specifying the GROUPBY=ROW plot-option.

The TWOWAY=GROUPHORIZONTAL layout produces a grouped plot that has a horizontal common baseline. By default (GROUPBY=COLUMN), the plot displays a block on the X axis for each column variable level. Within each column-level block, the plot displays row variable levels.

The TWOWAY=STACKED layout produces stacked displays of frequencies. By default (GROUPBY=COLUMN) in a stacked bar chart, the bars correspond to column variable levels, and row levels are stacked within each column level. By default in a stacked dot plot, the dotted lines correspond to column levels, and cell frequencies are plotted as data dots on the corresponding column line. The dot color identifies the row level.

The TWOWAY=CLUSTER layout, which is available only for bar charts, displays groups of adjacent bars. By default, the primary grouping is by column variable level, and row levels are displayed within each column level.

You can reverse the default row and column grouping in any layout by specifying the GROUPBY=ROW plot-option. The default is GROUPBY=COLUMN, which groups first by column variable.

TYPE=BARCHART | DOTPLOT

specifies the plot type (format) of the frequency (FREQPLOT), cumulative frequency (CUMFREQPLOT), and deviation plots (DEVIATIONPLOT). TYPE=BARCHART produces a bar chart and TYPE=DOTPLOT produces a dot plot. The default is TYPE=BARCHART.

PRINTKWTS

displays the agreement weights that PROC FREQ uses to compute the weighted kappa coefficient. Agreement weights reflect the relative agreement between pairs of variable levels. By default, PROC FREQ uses the Cicchetti-Allison form of agreement weights. If you specify the AGREE(WT=FC) option, the procedure uses the Fleiss-Cohen form of agreement weights. For more information, see the section Weighted Kappa Coefficient.

This option has no effect unless you also specify the AGREE option to compute the weighted kappa coefficient. The PRINTKWTS option is equivalent to the AGREE(PRINTKWTS) option.

RELRISK OR

requests relative risk measures and their asymptotic Wald confidence limits for $2 \times 2$ tables. These measures include the odds ratio, the column 1 relative risk, and the column 2 relative risk. PROC FREQ displays the statistics and confidence limits in the “Odds Ratio and Relative Risks” table. For more information, see the section Odds Ratio and Relative Risks for 2 x 2 Tables.

You can request exact confidence limits for the odds ratio by specifying the OR option in the EXACT statement. You can request exact unconditional confidence limits for the relative risks by specifying the RELRISK option in the EXACT statement. For more information, see the sections Exact Confidence Limits for the Odds Ratio and Exact Unconditional Confidence Limits for the Relative Risk.

You can request score confidence limits for the odds ratio by specifying the OR(CL=) option. This option produces the “Confidence Limits for the Odds Ratio” table, which can also display Wald and exact confidence limits. For more information, see the section Score Confidence Limits for the Odds Ratio.

You can also obtain the relative risk measures by specifying the MEASURES option, which produces other measures of association in addition to the relative risks.

RISKDIFF < (riskdiff-options)>

requests risks (binomial proportions) and risk differences for $2 \times 2$ tables. When you specify the RISKDIFF option, PROC FREQ provides the row 1 risk, row 2 risk, total (overall) risk, and risk difference (row 1 – row 2), together with their asymptotic standard errors and Wald confidence limits. PROC FREQ also provides exact (Clopper-Pearson) confidence limits for the row 1, row 2, and total risks by default. You can request exact unconditional confidence limits for the risk difference by specifying the RISKDIFF option in the EXACT statement. For more information, see the section Risks and Risk Differences. PROC FREQ displays these results in the column 1 and column 2 “Risk Estimates” tables.

You can specify riskdiff-options in parentheses after the RISKDIFF option to request tests and additional confidence limits for the risk difference, as well as estimates of the common risk difference for multiway $2 \times 2$ tables. Table 3.18 summarizes the riskdiff-options.

The CL= riskdiff-option requests confidence limits for the risk difference. Available confidence limit types include Agresti-Caffo, exact unconditional, Hauck-Anderson, Miettinen-Nurminen (score), Newcombe, and Wald. Continuity-corrected Newcombe and Wald confidence limits are also available. You can request more than one type of confidence limits in the same analysis. PROC FREQ displays the confidence limits in the “Confidence Limits for the Proportion (Risk) Difference” table.

The CL=EXACT riskdiff-option displays exact unconditional confidence limits in the “Proportion (Risk) Difference Confidence Limits” table. When you use CL=EXACT, you must also request computation of the exact confidence limits by specifying the RISKDIFF option in the EXACT statement.

The EQUIV, NONINF, and SUP riskdiff-options request tests of equivalence, noninferiority, and superiority, respectively, for the risk difference. Available test methods include Farrington-Manning (score), Hauck-Anderson, and Newcombe (hybrid-score), in addition to the Wald test.

As part of the noninferiority, superiority, and equivalence analyses, PROC FREQ provides null-based equivalence limits that have a confidence coefficient of $100(1-2\alpha )$ % (Schuirmann, 1999). The ALPHA= option determines the confidence level, and the default of ALPHA=0.05 produces 90% equivalence limits. See the sections Noninferiority Tests and Equivalence Tests for details.

Table 3.18: RISKDIFF (Proportion Difference) Options

Option	Description
COLUMN=1 \| 2	Specifies the risk column
COMMON	Requests common risk difference
CORRECT	Requests continuity correction
NORISKS	Suppresses default risk tables
Request Confidence Limits
CL=AC	Requests Agresti-Caffo confidence limits
CL=EXACT	Displays exact confidence limits
CL=HA	Requests Hauck-Anderson confidence limits
CL=MN \| SCORE	Requests Miettinen-Nurminen confidence limits
CL=NEWCOMBE	Requests Newcombe confidence limits
CL=WALD	Requests Wald confidence limits
Request Tests
EQUAL	Requests an equality test
EQUIV \| EQUIVALENCE	Requests an equivalence test
NONINF \| NONINFERIORITY	Requests a noninferiority test
SUP \| SUPERIORITY	Requests a superiority test
MARGIN=	Specifies the test margin
METHOD=	Specifies the test method
VAR=SAMPLE \| NULL	Specifies the test variance

You can specify the following riskdiff-options in parentheses after the RISKDIFF option:

CL=type | (types)

requests confidence limits for the risk difference. You can specify one or more types of confidence limits. When you specify only one type, you can omit the parentheses around the request. PROC FREQ displays the confidence limits in the “Proportion (Risk) Difference Confidence Limits” table.

The ALPHA= option determines the confidence level, and the default of ALPHA=0.05 produces 95% confidence limits for the risk difference. This differs from the test-based confidence limits that are provided with the equivalence, noninferiority, and superiority tests (EQUIV, NONINF, and SUP), which have a confidence coefficient of $100(1-2\alpha )$ % (Schuirmann, 1999).

You can specify CL= with or without requests for risk difference tests. The confidence limits produced by CL= do not depend on the tests that you request and do not use the value of the test margin (which you specify by using MARGIN= riskdiff-option).

You can control the risk column for the confidence limits by specifying the COLUMN= riskdiff-option. If you do not specify COLUMN=, PROC FREQ provides confidence limits for the column 1 risk difference by default.

The following types of confidence limits are available:

AGRESTICAFFO AC

requests Agresti-Caffo confidence limits for the risk difference. See the subsection “Agresti-Caffo Confidence Limits” in the section Risk Difference Confidence Limits for details.

EXACT

displays exact unconditional confidence limits for the risk difference in the “Proportion (Risk) Difference Confidence Limits” table. You must also request computation of the exact confidence limits by specifying the RISKDIFF option in the EXACT statement.

PROC FREQ computes the confidence limits by inverting two separate one-sided exact tests (tail method), where the tests are based on the unstandardized risk difference by default. If you specify the RISKDIFF(METHOD=SCORE) option in the EXACT statement, the tests are based on the score statistic. See the RISKDIFF option in the EXACT statement and the section Exact Unconditional Confidence Limits for the Risk Difference for more information.

By default, PROC FREQ also displays these exact confidence limits in the “Risk Estimates” table. You can suppress this table by specifying the NORISKS riskdiff-option.

HA

requests Hauck-Anderson confidence limits for the risk difference. See the subsection “Hauck-Anderson Confidence Limits” in the section Risk Difference Confidence Limits for details.

MN < (CORRECT=NO | MEE)> SCORE < (CORRECT=NO | MEE)>

requests Miettinen-Nurminen (score) confidence limits for the risk difference. For more information, see the subsection “Miettinen-Nurminen (Score) Confidence Limits” in the section Risk Difference Confidence Limits. By default, the Miettinen-Nurminen confidence limits include a bias correction factor (Miettinen and Nurminen, 1985; Newcombe and Nurminen, 2011). If you specify CL=MN(CORRECT=NO), PROC FREQ provides the uncorrected form of the confidence limits (Mee, 1984).

NEWCOMBE < (CORRECT)>

requests Newcombe hybrid-score confidence limits for the risk difference. If you specify CL=NEWCOMBE(CORRECT) or the CORRECT riskdiff-option, the Newcombe confidence limits include a continuity correction. See the subsection “Newcombe Confidence Limits” in the section Risk Difference Confidence Limits for details.

WALD < (CORRECT)>

requests Wald confidence limits for the risk difference. If you specify CL=WALD(CORRECT) or the CORRECT riskdiff-option, the Wald confidence limits include a continuity correction. See the subsection “Wald Confidence Limits” in the section Risk Difference Confidence Limits for details.

COLUMN=1 | 2 | BOTH

specifies the table column for which to compute the risk difference tests (EQUAL, EQUIV, NONINF, and SUP) and the risk difference confidence limits (which are requested by the CL= riskdiff-option).

If you do not specify COLUMN=, PROC FREQ provides the risk difference tests and confidence limits for column 1 by default. The COLUMN= option has no effect on the “Risk Estimates” table, which is produced for both column 1 and column 2. You can suppress the “Risk Estimates” table by specifying the NORISKS riskdiff-option.

COMMON

requests estimates of the common (overall) risk difference for multiway $2 \times 2$ tables. PROC FREQ produces Mantel-Haenszel and summary score estimates for the common risk difference, together with their confidence limits. For more information, see the section Common Risk Difference.

If you do not specify the COLUMN= riskdiff-option, PROC FREQ provides the common risk difference for column 1 by default. If you specify COLUMN=2, PROC FREQ provides the common risk difference for column 2. COLUMN=BOTH does not apply to the common risk difference.

CORRECT

includes a continuity correction in the Wald confidence limits, Wald tests, and Newcombe confidence limits. See the section Risks and Risk Differences for details.

EQUAL

requests a test of the null hypothesis that the risk difference equals zero. PROC FREQ provides an asymptotic Wald test of equality. If you specify the CORRECT riskdiff-option, the Wald test includes a continuity correction. If you specify the VAR=NULL riskdiff-option, the test uses the null (test-based) variance instead of the sample variance. See the section Equality Test for details.

EQUIV EQUIVALENCE

requests a test of equivalence for the risk difference. See the section Equivalence Tests for details. You can specify the equivalence test margins with the MARGIN= riskdiff-option and the test method with the METHOD= riskdiff-option. PROC FREQ uses METHOD=WALD by default.

MARGIN=value | (lower, upper)

specifies the margin for the noninferiority, superiority, and equivalence tests, which you request with the NONINF, SUP, and EQUIV riskdiff-options, respectively. If you do not specify MARGIN=, PROC FREQ uses a margin of 0.2 by default.

For noninferiority and superiority tests, specify a single value for MARGIN=. The MARGIN= value must be a positive number. You can specify value as a number between 0 and 1. Or you can specify value in percentage form as a number between 1 and 100, and PROC FREQ converts that number to a proportion. The procedure treats the value 1 as 1%.

METHOD=method

specifies the method for the noninferiority, superiority, and equivalence analyses, which you request with the NONINF, SUP, and EQUIV riskdiff-options, respectively. If you do not specify the METHOD= riskdiff-option, PROC FREQ uses METHOD=WALD by default.

The following methods are available:

FM SCORE: requests Farrington-Manning (score) tests and equivalence limits for the equivalence, noninferiority, and superiority analyses. See the subsection “Farrington-Manning (Score) Test” in the section Noninferiority Tests for details.
HA: requests Hauck-Anderson tests and confidence limits for the equivalence, noninferiority, and superiority analyses. See the subsection “Hauck-Anderson Test” in the section Noninferiority Tests for details.
NEWCOMBE: requests Newcombe (hybrid-score) confidence limits for the equivalence, noninferiority, and superiority analyses. If you specify the CORRECT riskdiff-option, the Newcombe confidence limits include a continuity correction. See the subsection “Newcombe Noninferiority Analysis” in the section Noninferiority Tests for details.
WALD: requests Wald tests and confidence limits for the equivalence, noninferiority, and superiority analyses. If you specify the CORRECT riskdiff-option, the Wald confidence limits include a continuity correction. If you specify the VAR=NULL riskdiff-option, the tests and confidence limits use the null (test-based) variance instead of the sample variance. See the subsection “Wald Test” in the section Noninferiority Tests for details.

NONINF NONINFERIORITY

requests a test of noninferiority for the risk difference. See the section Noninferiority Tests for details. You can specify the test margin with the MARGIN= riskdiff-option and the test method with the METHOD= riskdiff-option. PROC FREQ uses METHOD=WALD by default.

NORISKS

suppresses display of the “Risk Estimates” tables, which the RISKDIFF option produces by default for column 1 and column 2. The “Risk Estimates” tables contain the risks and risk differences, together with their asymptotic standard errors, Wald confidence limits, and exact confidence limits.

SUP SUPERIORITY

requests a test of superiority for the binomial proportion. See the section Superiority Test for details. You can specify the test margin with the MARGIN= riskdiff-option and the test method with the METHOD= riskdiff-option. PROC FREQ uses METHOD=WALD by default.

VAR=SAMPLE | NULL

specifies the type of variance estimate to use in the Wald tests of noninferiority, superiority, equivalence, and equality. The default is VAR=SAMPLE, which uses the sample variance. VAR=NULL uses a test-based variance that is computed by using the null hypothesis risk difference value. See the sections Equality Test and Noninferiority Tests for details.

SCORES=type

specifies the type of row and column scores that PROC FREQ uses to compute the following statistics: Mantel-Haenszel chi-square, Pearson correlation, Cochran-Armitage test for trend, weighted kappa coefficient, and Cochran-Mantel-Haenszel statistics. The value of type can be one of the following:

MODRIDIT
RANK
RIDIT
TABLE

See the section Scores for descriptions of these score types.

If you do not specify the SCORES= option, PROC FREQ uses SCORES=TABLE by default. For character variables, the row and column TABLE scores are the row and column numbers. That is, the TABLE score is 1 for row 1, 2 for row 2, and so on. For numeric variables, the row and column TABLE scores equal the variable values. See the section Scores for details. Using MODRIDIT, RANK, or RIDIT scores yields nonparametric analyses.

You can use the SCOROUT option to display the row and column scores.

SCOROUT

displays the row and column scores that PROC FREQ uses to compute score-based tests and statistics. You can specify the score type with the SCORES= option. See the section Scores for details.

The scores are computed and displayed only when PROC FREQ computes statistics for two-way tables. You can use ODS to store the scores in an output data set. See the section ODS Table Names for more information.

SPARSE

reports all possible combinations of the variable values for an n-way table when n > 1, even if a combination does not occur in the data. The SPARSE option applies only to crosstabulation tables displayed in LIST format and to the OUT= output data set. If you do not use the LIST or OUT= option, the SPARSE option has no effect.

When you specify the SPARSE and LIST options, PROC FREQ displays all combinations of variable values in the table listing, including those with a frequency count of zero. By default, without the SPARSE option, PROC FREQ does not display zero-frequency levels in LIST output. When you use the SPARSE and OUT= options, PROC FREQ includes empty crosstabulation table cells in the output data set. By default, PROC FREQ does not include zero-frequency table cells in the output data set.

See the section Missing Values for more information.

TOTPCT

displays the percentage of the total multiway table frequency in crosstabulation tables for n-way tables, where n > 2. By default, PROC FREQ displays the percentage of the individual two-way table frequency but does not display the percentage of the total frequency for multiway crosstabulation tables. See the section Two-Way and Multiway Tables for more information.

The percentage of total multiway table frequency is displayed by default when you specify the LIST option. It is also provided by default in the PERCENT variable in the OUT= output data set.

TREND

requests the Cochran-Armitage test for trend. The table must be $2 \times C$ or $R \times 2$ to compute the trend test. See the section Cochran-Armitage Test for Trend for details. To request exact p-values for the trend test, specify the TREND option in the EXACT statement. See the section Exact Statistics for more information.