# Analysis of Repeated Measures Study with Unequal Time Between Measurements

I have inherited the analysis of a study from a previous research student. I am having trouble determining what type of analysis would be most appropriate to use.

I am trying to assess whether or not a significant change in measurement scores (FMD) occurred comparing before and after an intervention.

~25 participants were measured at Time 1, then re-tested under near identical circumstances 1 week later at Time 2. Then a 6 month intervention was implemented and the participants were re-tested after the intervention (at Time 3 and Time 4). Time 3 and Time 4 were also conducted under near identical circumstances 1 week apart.

I have been struggling to find an appropriate way to analyze all of these measurement points while also attributing the dependence between them appropriately.

I do not feel that a repeated measures ANOVA is appropriate as it will assume that all the measurements are equally dependent (which is not the case - Time 1 vs Time 2 are more related than Time 1 vs Times 3 or 4)

I have considered using a Multilevel Linear Model, however I am unfamiliar with the process and cannot find an appropriate covariance structure in the predefined SPSS options (I also know that SPSS is not the best program for multi-level modelling).

My understanding of statistical analysis and theory is fairly limited (not a stats or extensive research background). If Multilevel Modelling is the way to go, is there a way to model this scenario appropriately in SPSS or would I have to learn another program? Any advice would be deeply appreciated.

Thanks,

Rafreaki

Not being an expert, but the repeated measure ANOVA assumption you are stating seems confusing to me:

I do not feel that a repeated measures ANOVA is appropriate as it will assume that all the measurements are equally dependent (which is not the case - Time 1 vs Time 2 are more related than Time 1 vs Times 3 or 4

Dependent on what? Perhaps you mean the assumption of sphericity? Sphericity refers to the equality of variances of the differences between treatment levels. (Field, 2010, p395) Have a look at Field - he explains it really well. This SE question also talks about the sphericity assumption and there are many more on SE.

Sphericity can be checked with Mauchly's test. The commonly used statistics program like SPSS and SAS compute it automatically for you.

• (+1) but having recently read that Field article, I can't recommend it. Jun 6, 2018 at 19:45
• What I am referecing is not Field's article but rather his book. I've added a link to avoid confusion in future. I guess what you are hinting at is linear mixed (multilevel) models? Jun 7, 2018 at 11:49
• Where I saw problems was discoveringstatistics.com/docs/repeatedmeasures.pdf. But on returning I don't see many :-) Chiefly the confusion of Type I and Type II error. Jun 7, 2018 at 13:36

interrupted time series analysis? too few timepoints to get a sense of the trend before and after the intervention. Otherwise this is a standard approach for such data. I could dig out a reference if you need it but i'm not familiar with spss