Hello,
I am using -cc- in Stata 13.
I use the -cc- command with the by() option to get...
Test of homogeneity;
Test that combined OR =1, yielding a Mantel-Haenszel result.
Both of these are chi2 results, even with -exact- and -bd- (Breslow-Day) option.
A colleague also ran the corresponding tests in Cytel Studio, and obtained corresponding results under the heading "Mantel-Haenszel Inference".
One difference is that, in some cases, Cytel Studio uses a DF value that is 1 less than that used in Stata-cc-.This happens when one of the
strata yields missing results. My colleague believes that the DF in Cytel Studio is correct in these cases; the appropriate action is to discount that category. That is, DF should be num_strata-1, where num_strata counts only those strata that contribute to the analysis.
Cytel Studio also yields what is denoted an Exact result:
Exact Estimation of Common Odds Ratio
Exact p-values for testing that the Common odds Ratio is 1
Why does Stata's -cc- yield only the Mantel-Haenszel inference, and not the exact estimation? (The two types of results often differ. Furthermore sometimes -cc- yields a missing value, where Cytel Studio yields an actual value.) Does anyone know how to compute the Exact result -- to either enhance -cc- or create a separate program? Or is there already a program available to do this?
I ask this because I find the Exact result is often preferable. Also, I need to iterate my analysis over many subsets, and may also need to adjust the specifications and rerun. The process of transferring data over to Cytel Studio and running analyses there is laborious. We also want to capture the results, which I know how to do in Stata; I'm not sure if it is possible in Cytel Studio.
Thanks to anyone who can help.
--David
I am using -cc- in Stata 13.
I use the -cc- command with the by() option to get...
Test of homogeneity;
Test that combined OR =1, yielding a Mantel-Haenszel result.
Both of these are chi2 results, even with -exact- and -bd- (Breslow-Day) option.
A colleague also ran the corresponding tests in Cytel Studio, and obtained corresponding results under the heading "Mantel-Haenszel Inference".
One difference is that, in some cases, Cytel Studio uses a DF value that is 1 less than that used in Stata-cc-.This happens when one of the
strata yields missing results. My colleague believes that the DF in Cytel Studio is correct in these cases; the appropriate action is to discount that category. That is, DF should be num_strata-1, where num_strata counts only those strata that contribute to the analysis.
Cytel Studio also yields what is denoted an Exact result:
Exact Estimation of Common Odds Ratio
Exact p-values for testing that the Common odds Ratio is 1
Why does Stata's -cc- yield only the Mantel-Haenszel inference, and not the exact estimation? (The two types of results often differ. Furthermore sometimes -cc- yields a missing value, where Cytel Studio yields an actual value.) Does anyone know how to compute the Exact result -- to either enhance -cc- or create a separate program? Or is there already a program available to do this?
I ask this because I find the Exact result is often preferable. Also, I need to iterate my analysis over many subsets, and may also need to adjust the specifications and rerun. The process of transferring data over to Cytel Studio and running analyses there is laborious. We also want to capture the results, which I know how to do in Stata; I'm not sure if it is possible in Cytel Studio.
Thanks to anyone who can help.
--David
Comment