02-03-2015 05:05 PM
I am working on a cross table, for instance for the cross table Hospital regions vs ARF-NoARF hospitalizations, it would be 2 x 4 cross table and the p value yielded is applicable to the whole tables combined. How do I get the p values for difference between ARF-NOARF admissions for each region separately, the poct-hoc comparisons?
I am attaching a part of the table and its figure here to this post.I want the p values for the highlighted #s in the table. I have also plotted them on the graph next to the table. All I need is the p values for the red/blue bars for each region.
The overall p value that’s shown in the table is the combined p value. But in the graph you see that the ARF and NIS %s are very different for northeast and South regions but NOT very different for Midwest and West regions. So how do I get those p values exactly, separate for each region from this table that would allow me say if they are significant or not?
SO basically the code I have used is below and that’s how I got the table below and I created the graph. You can take a look at the Figure 3 in that paper and read the description right below it.
table arf hospitalization *region /chisq;
Any help would be greatly appreciated!
02-05-2015 04:49 PM
Thanks very much Steve.! I tried the above code by using another example.Regionwise afib hospitalizations. I am attaching the output to the original posts. I am not sure how to interprete the results though.
For this analyses, I am referring to a published article and I've highlighted the part that I am trying to do for my analyses.In that paper, they have mentioned the posthoc comparisons in 3 figures which I have highlighted in the article and attached it to the oiginal post. I have made similar graphs and and my results are quite similar to theirs so I would like to do the posthoc comparisons as well. I have also attached my own results in the original post.
Please take a look and if yo could advise on how I can obtain those p values; for the post-hoc comparisons ,that would bevery helpful!
02-06-2015 08:20 AM
Well, for each region you get: proportion in each category, the asymptotic standard error, confidence bounds, and a test of whether the proportion is equal to 0.5, which is equivalent to asking whether the categorization yields a different proportion than simple randomization.
So the procedure in the paper most likely did the following:
If that is significant, then follow-up with pairwise comparisons:
where region=1 or region=2;/* This coding will depend on how regions are coded*/
and then stepping through all pairwise comparisons of region. If there are a lot of comparisons, you may need to consider passing the raw p values to PROC MULTTEST to get adjusted p values, but it does not appear that Miyake et al. did that.