Hello experts!
I was just struggling with what kind analysis I might carry out for a certain dataset.
I have a sample of 200 cases (IDs) divided into three different groups; I would like to analyze the scores of each intervention (score having taken a pre-assessment, score taking some pre-assessment, and score with no pre-assessment). Initially, I was thinking of using an ANOVA, however the dataset violates more than a few assumptions (e.g., the subjects have duplicate scores, some subjects are in all three groups multiple times, no homoscedasticity, data were not randomly sampled).
So since I just want to analyze whether having a degree intervention influences scores, I figure an ANOVA would be appropriate, and I would not be looking for between-subject/group effects.
Would I be okay in carrying out this kind of analysis if I'm simply looking at whether having full/some/no pre-assessment has an overall impact on their final grade in a class?
Thank you in advance for the help!
Dear Darko,
I would first recommend removing repeated subjects from your data set before any analysis.
Perhaps the best choice would then be the nonparametric Jonckheere-Terpstra test as you appear to be looking for a trend between your three groups. It is available in SPSS. If you get a significant result you can double click on the output table to open the Model Viewer which will then allow you to compare groups pairwise.
Hello,
I belive that after you remove the repeated subjects you can use MANOVA.
I hope I understood your need.
KEEP in touch.
When applying ANOVA, beside the other assumptions, the first one is that the samples are related, which is not your case.
For comparison between independent samples, you could apply the nonparametric Jonckheere-Terpstra test or the Kruskal-Wallis H test.
You can take a look here:
https://statistics.laerd.com/features-selecting-tests.php
Good luck.
You should be considering a mixed-effects model. Treat your ID and timepoint readings as random effects such as (Timepoint|ID), where timepoint is a factor with preassessement and post-assessment levels ( 2 codes) and ID a factor with 200 levels. The random-effects term will produce an intercept and slope model for the random effects mentioned. In R, lmer can handle the imbalance and missing values you mentioned. Just test the residuals of the model via a qqplot to check overall goodness of fit. In R, the model would look something like
Response ~ Group + (Timepoint|ID).
If your interested also in the difference in pre and postment then use something like:
Response ~ Group*Timepoint + (Timepoint|ID)
Group and Timepoint should be fully crossed factors. Anova can be used on the resulting model to test the significance of the effects for either model given above.
Have a look at the following - where you can use anova depending your needs.
Main Assumptions for ANOVA -
1. The population in which samples are drawn should be normally distributed.
2. Independence of cases: the sample cases should be independent of each other.
3. Homogeneity: Homogeneity means that the variance between the groups should be approximately equal.
Hope you will find your desired answer.
Thanks
http://www.statisticssolutions.com/manova-analysis-anova/
http://support.minitab.com/en-us/minitab/17/topic-library/modeling-statistics/anova/basics/what-is-anova/
http://onlinestatbook.com/2/analysis_of_variance/intro.html
- Badges
- Science topic