topic Re: Selecting the right statistical test in Statistical Procedures

Selecting the right statistical test

duckypooh — Mon, 22 Jun 2020 10:56:22 GMT

Hello!

Sorry if my question seem basic but i am new to statistical analysis and would to ask which is the best test to determine association between variables and outcome if:

1. I have a small sample size ~15

2. 1 binary independent variable

3. A mixture of continuous and categorical dependent variables

TIA!

Re: Selecting the right statistical test

PaigeMiller — Mon, 22 Jun 2020 11:38:00 GMT

With a sample size of ~15, no test will have a lot of power.

For continuous dependent variables, I would use PROC GLM Analysis of Variance (assuming the errors are iid normal), and PROC NPAR1WAY otherwise.

For categorical variables, I would use a contingency table in PROC FREQ and do the Chi-Squared test.

Re: Selecting the right statistical test

Ksharp — Mon, 22 Jun 2020 11:40:53 GMT

You didn't post your data yet ?
1) Fisher Exact Test ?
proc freq;
table a*b/chisq fisher;
exact fisher;
run;

2) Logistic Regression ?
proc logistic;
model y=

3) General Linear Model ?
proc glm

Re: Selecting the right statistical test

SteveDenham — Mon, 22 Jun 2020 12:39:26 GMT

One more choice, based on the small sample size, would be to use the EXACT option in PROC LOGISTIC. However (and this is critical), you need to look at the cross tabulation of your results versus the categorical variables. With this small sample size, you are almost certain to have what is called complete or partial separation. I would suggest removing from the analysis any variables that show complete separation - you know what the outcome will be once you have the value of the independent categorical variable. If you still run into partial separation issues, then you might consider using PROC GENMOD and the alternating logistic regression method outlined here0

:https://documentation.sas.com/?docsetId=statug&docsetTarget=statug_genmod_examples06.htm&docsetVersion=15.1&locale=en

SteveDenham

Re: Selecting the right statistical test

Ksharp — Mon, 22 Jun 2020 12:52:14 GMT

Agree. That is why I said " did not post the data yet "

Re: Selecting the right statistical test

Ksharp — Mon, 22 Jun 2020 12:53:43 GMT

One more thing,
OP said ASSOCIATION not CORRELATION , Therefore that is bring me into contingency table analysis .

Re: Selecting the right statistical test

SteveDenham — Mon, 22 Jun 2020 14:20:34 GMT

Good point @Ksharp .for the categorical variables. I guess I got caught up in the continuous side, where modeling makes some sense, even with a sample size as small as this.

SteveDenham

Re: Selecting the right statistical test

ballardw — Mon, 22 Jun 2020 14:47:29 GMT

@duckypooh wrote:

Hello!

Sorry if my question seem basic but i am new to statistical analysis and would to ask which is the best test to determine association between variables and outcome if:

1. I have a small sample size ~15

2. 1 binary independent variable

3. A mixture of continuous and categorical dependent variables

TIA!

What type of research question is to be answered? Question will point towards choice of test(s).

Are your categorical variables possibly ordinal? That means an order could be determined "Bad neutral Good" as apposed to simple category " Fred Orange XYZZY"?