How does one get around performing a 2 Proportion T-Test in SPSS (A dichotomous variable is being measured between two groups)? Doing a Chi-Squared will not provide the mean difference with either a STD or 95% CI associated with it, which I need.
Hello Jack,
The Fisher exact test will make this comparison. In SPSS, from the main menu bar: Analyze / Descriptive statistics / Crosstabs...
Groups will be one dimension of your table, whereas status (in target category, not in target category) will be the other. Click the Statistics button to check off the chi-square test. (In older versions, you might also have to check the "Risk" box in this dialog.) Among the outputs will be the results of the Fisher exact test (directional and nondirectional).
Good luck with your work.
Thank you David,
Will the Fisher exact test give an output of the difference between the proportions with associated standard deviation or 95% confidence interval as well? Ex: Mean proportion Group (a), Mean proportion Group (b), Difference (a-b) +/- STD or 95% CI.
Jack Harris, it sounds to me like you want a test on the risk difference. If your dependent variable is coded 1=Yes, 0=No for the event of interest, then use GENLIN with the following settings on the /MODEL sub-command:
- DISTRIBUTION=NORMAL LINK=IDENTITY
...and these settings on the /CRITERIA subcommand:
- SCALE=MLE COVB=ROBUST
With 1/0 coding as above, the means = proportions. Use EMMEANS sub-commands as desired to display proportions and differences in proportions. See documentation for GENLIN here:
https://www.ibm.com/support/knowledgecenter/en/SSLVMB_26.0.0/statistics_reference_project_ddita/spss/advanced/syn_genlin.html
Here's an example you may be able to adapt:
* Generate a sample dataset--replace N values with your own.
NEW FILE.
DATASET CLOSE ALL.
DATA LIST LIST / Group Event N (3F5.0).
BEGIN DATA
1 1 50
1 0 450
2 1 100
2 0 400
END DATA.
WEIGHT by N.
GENLIN Event BY Group
/MODEL Group INTERCEPT=YES
DISTRIBUTION=NORMAL LINK=IDENTITY
/CRITERIA SCALE=MLE COVB=ROBUST ANALYSISTYPE=3(LR)
CILEVEL=95 CITYPE=WALD LIKELIHOOD=FULL
/MISSING CLASSMISSING=EXCLUDE
/EMMEANS TABLES=Group SCALE=ORIGINAL COMPARE=Group CONTRAST=DIFFERENCE PADJUST=LSD
/PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION
.
HTH.
EDIT
PS- For validation, I just tried the following using the csi (i.e., cohort-study-immediate) in Stata:
csi 50 100 450 400
It shows a risk difference of -0.10, 95% CI, -.1438261 to -.0561739. When I increase the number of decimals in the SPSS output from GENLIN, the 95% CI for the Group coefficient matches to 7 decimals the CI from Stata.
- Badges
- Science topic