I would very much welcome your expertise in choosing a route with my data, raw using Excel then on SPSS 26.
I have carried out an online questionnaire for my MSci project. All factors have 3 timepoint responses. The research question is to identify if 3 within-subject timepoints (IV) affect scores of mental health Likert Scale scores (4 within-subject DVs classified as continuous, as no other repeated measures test available to identify if score are affected over time). Also, to identify if mental health scores interact with level of activity intensity, another within-subject categorical factor (2 or 4 categories). Due to having 3 timepoints, I anticipated using repeated-measures ANOVA, however, the intensity of activity W-S factor can have 2 or 4 levels - the 2 level option has 41 participants in one group and 120 in the other. With 4 levels only 10, 31, 53, 67 in each.
Is the only way to analyse to meet the assumptions, to ditch 2/3rds of 120 group data and randomly select 41 of them, so there are equal groups to carry out RM ANOVA? Hardly worth reducing the 4 levels to groups to 10? Or is it better to do individual repeated-measure ANOVAs on each DV and the intensity of activity variable and then if any significance found run further analyses with multiple regression?
There are a fair few outliers and due to Likert Scale responses, zero answers are skewing the distribution curves.
I've done the classic not enough time spent on identifying analysis tests prior to running the questionnaire (due to COVID exceptional circumstances). Can you help?
Hello Sharon,
I'm not completely clear on what your intended analysis entails, but can offer some initial suggestions.
1. Do not jettison cases from a larger group/condition so that the per-group sizes are equal. It is true that the "effective" per group size, as regards how n relates to statistical power tends to track the harmonic mean of group sample sizes rather than the arithmetic mean. But that's even more of a reason not to toss out presumably perfectly good data.
2. I wasn't sure of what a "zero score" to an item meant: If it is a non-response, then that's quite different from selecting a response of "Not applicable" or that of the lowest value of the Likert-type response scale (assuming it is given a value of 0). The options of how to treat each are quite different.
3. Yes, it sounds as if the basic framework could be RM anova. It would be up to you to convince your examining committee that treating the scores as continuous made sense (vs. merely allowing you to use the parametric procedure). There are non-parametric analogues (such as the aligned ranks method: see this link for a quick introduction: https://rcompanion.org/handbook/F_16.html). As well, so-called hybrid techniques have been around for decades (Puri & Sen, Iman & Conover are among the best known). If distributional concerns are the matter that have your attention, then resampling/bootstrap methods are becoming quite common as options in many statistical software packages. Using such an approach allows you to proceed without concern for the shape of score distributions.
Apologies if I've misunderstood your query.
Good luck with your work.
Thank you David Morse for your prompt reply. The questions are separated into 3 time-points in the questionnaire. At each time-point, the participant was asked their activity level and mental health level, providing 5 DV's (4 Likert Scale - continuous, to use parametric as you stated) and 1 categorical variable with either 2 or 4 levels - health benefits or no health benefits. Aiming to establish if lower mental health scores are related to higher intensity exercise at the 3 time-points.
1. Thankful, as I have 160+ responses.
2. Zero score relates to, for example, not suffering from any anxiety symptoms whereas the next score may be 2 for some anxiety symptoms.
3. My supervisor and I have discussed all reasoning that has led to final analysis, backed by studies. Just seems evident that every example of factorial or two-way RM ANOVA has equal group membership prior to analysis.
4. Only using methodology Field's SPSS book as advised by my Student Services. Hence, why I am feeling restricted to achieve what the RQ set out to do.
I am appreciative that you have given your time in assisting in this regard.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA