- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Hi all,
I have a dataset (see example below) of the number of social media posts by platform type and account type. I've run a Fisher's exact test which has turned out to be significant and now I want to determine where the actual differences in proportion lie. For example, I'd like to determine if the proportion of posts by Account 1 on Platform1 are significantly different in comparison to all other account types. I've performed a post hoc test where I made a new dichotomous variable, 1=Account 1, 0=all other accounts and then run a Fisher's exact on this 3x2 table.
EXAMPLE TABLE:
Frequency of Posts
Account1 | Account2 | Account3 | Account4 | Account5 | |
Platform1 | |||||
Platform2 | |||||
Platform3 |
QUESTIONS:
1) I'm wondering if based on my aims, it's correct to group the columns as I have for a post hoc test. Most examples of pairwise comparisons as post hoc tests that I've seen tend to group the rows rather than the columns.
EXAMPLE PAIRWISE COMPARISON
Account1 | All other accounts | |
Platform1 | ||
Platform2 | ||
Platform3 |
Bonferroni p<0.0001
2) From my output, I get 5 p values (1 per each account type). For example, if the p value of my first pairwise comparison (see example above) is significant, then does this mean that there is a significant difference in the proportion of posts for Account1 compared to all other accounts for each of the 3 platforms? Is there a way in SAS to determine which exact cells within one row are significantly different between account types?
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
See this note. You might want to fit a Poisson model and then use LSMESTIMATE statements as shown in the log linear model section there to make the comparisons. If desired, you could then gather all of the p-values in a data set and then use PROC MULTTEST to do an adjustment for multiple testing as shown earlier in the note.