Nonparametric multivariate survey analysis - suggestions please!


New Member

I'm about to start analysis on some educational survey data with the following predictor and response variables:

PREDICTOR: age class (2 options), setting (5 options), experience (5 options)
RESPONSE: 30 survey questions, some Likert scale, some binary answers. These relate to two educational issues: how teachers plan their teaching, and how they assess it.

The number of respondents in each predictor class is unbalanced. There are 1400 respondents in total.

I would like to investigate the relative importance of each predictor on the way teachers plan teaching and assessment. So, for example, I would like to discover "that young teachers in preschools are significantly less confident in planning than teachers of other ages or in other settings"

My current thinking is that the best method would be to reduce the 30 questions to two scores, where each teacher has a score for planning, and another for assessment.

However, I'm currently unsure which multivariate analysis method is suitable once I've done this. I realise I have to run the test twice (once for planning, once for assessment), but which test?

Also, I currently have Minitab. Anyone use this? Or am I better off checking out SPSS? If so, any advice about how to write the model in SPSS/Minitab would be much appreciated!

Thanks in advance.


New Member
Yes you can create scores for assessment and planning. Multiple regression would be appropriate. The independent variables if numerical or dichotomous (i.e with 2 options) can be placed into the regression equation directly. You will have to do the test twice for each score. If the independent variables are categorical (three or more categorical response options), then you can be dummy-code (i.e. break the categorical variables to create dichotomous ones) and enter them into the equation


New Member
Thank you

Hi Dr D, thanks for this excellent advice. Now I know what to aim for.

Carrying this out in practice may be a little more difficult :) I'm going to play with the data this week and aim towards your suggestion. I hope I can work it out. I may need to ask for some more advice, but I should have a go first!