Dear all,
How can calculate Cohen's d in mixed linear models using interventional data with missing values? I want to compare the effects of an interventional trial. As usual, many participants show monoton missing data patterns.
Is there any helpful practical guide how to derive Cohen's d for cross group comparisons and within group changes in such situation? I used SPSS for my analysis.
I am really greatful for every good advise! Thanks in advance,
Best, Manuel
There is no simple answer, because it depends on why you want the effect size statistic. Cohen's d is defined as a population measure for an independent two group design. So one answer is to calculate a quantity that is comparable to that statistic. This would be to use the unstandardized coefficient for group (the estimate of differences in the means) and divide it by the standard deviation of the outcome in the raw data. This assumes that SD doesn't vary between the groups. If the groups have similar SDs then using the pooled group SD (as one would in the t test situation). If the groups do have different SDs then one can use a baseline or control condition SD (as per Cohen's original definition of d - though this measure is usually called Glass' D). If the SDs are very different and neither is an obvious baseline comparison then a d measure effect size might not be appropriate.
Furthermore, one could argue that standardized effect size is not that useful and the understandardized difference between the means is the more useful statistic.
There is no simple answer, because it depends on why you want the effect size statistic. Cohen's d is defined as a population measure for an independent two group design. So one answer is to calculate a quantity that is comparable to that statistic. This would be to use the unstandardized coefficient for group (the estimate of differences in the means) and divide it by the standard deviation of the outcome in the raw data. This assumes that SD doesn't vary between the groups. If the groups have similar SDs then using the pooled group SD (as one would in the t test situation). If the groups do have different SDs then one can use a baseline or control condition SD (as per Cohen's original definition of d - though this measure is usually called Glass' D). If the SDs are very different and neither is an obvious baseline comparison then a d measure effect size might not be appropriate.
Furthermore, one could argue that standardized effect size is not that useful and the understandardized difference between the means is the more useful statistic.
- Badges
- Science topic