topic Re: PROC SURVEYFREQ - Tests for trend in Statistical Procedures

PROC SURVEYFREQ - Tests for trend

_maldini_ — Mon, 21 Nov 2022 11:25:27 GMT

What options can be used to test of trend(s) in PROC SURVEYFREQ?

This LexJansen paper is great, but none of these options (Jonckheere-Terpstra, Cochran- Armitage and Cochran-Mantel-Haenszel tests) appear to be available w/ PROC SURVEYFREQ.

Thanks.

Re: PROC SURVEYFREQ - Tests for trend

ger15xxhcker — Sat, 10 Dec 2022 14:48:02 GMT

The PROC SURVEYFREQ procedure supports several options to test for trends in the data:

1. The CHISQ option, which performs a Pearson Chi-Square test to detect associations.

2. The SPEARMAN option, which performs Spearman rank correlation to test for monotonic trends.

3. The COCHRANQ option, which performs Cochran-Armitage trend test to detect linear trends.

4. The TTEST option, which performs a two-sample Student's t-test to test if means differ along with the specified ordering of the values.

5. The Wilcoxon option, which performs the Wilcoxon rank sum test to determine whether two trends diverge or are similar.

Re: PROC SURVEYFREQ - Tests for trend

_maldini_ — Wed, 28 Dec 2022 18:31:14 GMT

My understanding is that this option is for differences between categorical variables, not trends. Am I wrong?

Would this option be used to assess for a trend in ordinal categories of a dependent variable and an independent variable that has more than two ordered categories.?

If I was trying to determine a trend in quantiles of HDL cholesterol, for example, over time (i.e., years), what would be the proper SURVEY syntax?

Something like this?

PROC SURVEY LOGISTIC DATA=data_set;
TABLES hdl_quantiles*year / SPEARMAN;

My understanding is that this option is used to assess whether a trend is present between a binary (two levels, 0/1) dependent variable and an independent variable that has been categorized into more than two ordered categories.

If the non-SURVEY syntax were this (below):

proc freq data=chol9c;
table female*hdl_cat/trend noprint;
run;

What would the PROCSURVEY syntax look like?

Also, what SURVEY option should be used if the dependent variable is multinomial and not binary or ordinal (e.g., Race)? My understanding is that the Cochran-Mantel-Haenszel test would serve this purpose in a non-SURVEY procedure? The cmh option is not available in PROC SURVEYFREQ...

PROC FREQ data=dataset;
TABLES race*year/cmh noprint;
RUN;

Thanks.

Re: PROC SURVEYFREQ - Tests for trend

_maldini_ — Wed, 28 Dec 2022 18:49:55 GMT

What would be the SURVEY option for the "Jonckheere-Terpstra Test”, used to assess a trend between a non-normally distributed dependent variable and an ordered independent variable.?