What is the best statistical test/approach for pre- and post-intervention analysis?

arlp

New Member
#1
Dear colleagues,

We assessed several psychological measures (e.g. self-compassion scale, five-facets mindfulness questionnaire, satisfaction with life scale) before and after a Mindfulness-Based Psychoeducational Program for groups of people with acquired brain injuries. We would like to analyse whether these psychological variables improve significantly after the program. We tried to compare total scores of pre- and post- intervention, using paired-samples T-Tests and Wilcoxon Tests. Using these tests we had significant missing data due to attrition between time points. Is it possible to use another statistical tests, like One-Way ANOVA, in order to prevent losing data? Or would it be wrong?

Thank you very much
Best regards,
Raquel Pereira
 

Karabiner

TS Contributor
#2
Is this a pre-post design without untreated control group? How large is the sample/are the samples? How many missing data due to attrition are there?

With kind regards

Karabiner
 

arlp

New Member
#3
Dear Karabiner,

Thank you for your reply.

Yes, unfortunately that is a limitation: we don't have a control group. We have about 50 participants in total, which are the total sample collect through 9 months of running the treatment in 5/6 groups of +/- 10 people. Due to attrition, we loose about 20 participants.

My best regards,
Raquel
 

Karabiner

TS Contributor
#4
You can perform completer analyses. The research question for this would be whether pre-post values differ for those who completed the whole program.

In addition, you should inspect (by using descriptive statistics, in the first place, not significance tests) whether drop-outs and completers differed at baseline with regard to the psychological measures and other characteristics. This may give you information about generaliziability of findings, and maybe additional insight for whom the program is suited and for whom it is not.

In addition, you could perform some missing-value substitution. One easy (but critizised) way would be "last observation carried forward". Another easy (but critizised) way would be "hot deck imputation". If you want to use some elaborate method (which seems ethically be desirable in research with human patients), then have a look at "multiple imputation" technique.

With kind regards

Karabiner
 
Last edited: