I am evaluating a violence reduction strategy and have individual baseline, 6, 12 and 24 month follow-up data on the same people (obviously drops outs in there as well, so not complete data for all). What is the best statistical approach? I would like to be able to graph changes over time to see if we have a peak, plateau, etc.
The type of data may be important for how it is plotted. Does each individual have a single score? Are individual scores an average of several measures? Are several measures to be plotted separately?
A good method for visualizing change is to plot the data on a normal probability plot. The data need not be normal. It is the shape of the plot and how it changes with time. You can track indivduals for their position on the plot. You can also see how the distribution metrics change with time.
More details and suggestions are dependent upon the type of data.
Without knowing the exact nature of your data, I would suggest a repeated measures one-way ANOVA if you have a single "violence" score at each time point. And as far as graphing, a simple line graph should suffice for that type of data. You will lose subjects who have missing data, but that is the case with (almost) all longitudinal studies.
Edit: disregard what I said about spacing and regression, I was thinking of something else.
Now if you have multiple measures things become much more complicated, but that would require knowing more about the data you have.
If you have further questions you can email or direct message me.
Sandy,
latent change score models are able to explicitly model individual-level change over time. They can also accommodate unequal spacing between measurements.
For an introduction, see:
https://www.researchgate.net/publication/285651921_Longitudinal_Analysis_in_Occupational_Health_Psychology_A_Review_and_Tutorial_of_Three_Longitudinal_Modeling_Techniques
Article Longitudinal Analysis in Occupational Health Psychology: A R...
If the variable is interval and you are using each individual measure as its own observation, fixed effect models would be the simplest unbiased estimators. The longitudinal models are of course more advanced, but may require bigger sample sizes. YOu may want to check the literature in the area. At least you want to use the most sophisticated modeling technique in the literature or one that solves a specific problem with that technique.
As Heiko Breitsohl has suggested, latent change score mode should help you compare the mean scores over ti
Hierarchical models are suitable for repeated measures. But you have to examine the nature of your data to see if indeed they are the type that the model could be used for
ANOVA for repeated measures: for continuous data with required assumptions
Friedman Test: ordinal data or continuous data without required assumptions
Cochrane test: noninal (binomial) data
Hopefully, you can find answers herein.
Article StaTips Part I: Choosing statistical test when dealing with ...
Gibson
My colleagues who answered this question earlier e.g Claxton have proposed what I also believe can easily solve your problem.The ANOVA technique will help you detect if the means of violence reduce with the use of that strategy with time. The technique will therefore help you to argue that there is a significant (or not) difference in the levels of violence when the violence reduction strategy is used after 6, 12 or 24 months because it will get for you the mean violence levels for each baseline set of data. However, to use ANOVA, violence needs to be measured to scale i.e. at interval or ratio form. Do you get that? Good luck please!
Hi Sandy,
I support both David and Claxton because I used the the method they have given to you. Your study is similar to my PhD. I had Knowledge level, Attitudes changes and behavioural changes measures of assessments at pre-intervention, six months, one year, 18 months and 24 months. though I was dealing with primary school pupils, there were other factors that were included such as location of the school as some were in urban and others in the rural, the class levels etc, At each of these intervals some students were being lost like yours. Also there were insignificant changes after six months, therefore only 12 monthly measurement interval was adopted. I successfully used ANOVA repeated measure of the General Linear Model of SPSS. You can access my PhD thesis through university of keel Library. The title is The Impact of an HIV/AIDS Education Intervention Programme in Primary Schools in Malawi by Fletcher Mngongongonda Phiri, 2009. You might find it useful. You can come back to me if you still have questions after reading it. Best wishes
- Badges
- Science topic
- Similar topics
- Psychology
- Social Psychology