Mood: 
a-ok
Now Playing: Surfin’ USA (The Beach Boys)
Topic: Data Analysis
So, had my meeting with John this afternoon and James came along to the meeting as well. Well, I discussed with John my idea of converting the disciplines into soft and hard, pure and applied, life and non-life. He indicated that they were doing something similar in the the SOMUL project and that the person’s work they were using was Tony Becher for decided the disciplines (he only told me this since I was using Biglan’s categorization) – I think it might be similar or Becher using Biglan’s work for his classification. I remember the name Tony Becher though because I have a paper by Neumann et al which Becher co-authored concerning disciplines.
Anyway, James have some reservations of using this classification since he doesn’t think a lecturer think of themselves as being applied or pure at all -that name only comes because the department chooses to use the name applied in their title. Well … to some extent I think that is true – but that doesn’t prevent the shaping of course to reflect its applied nature and teachers conforming to that situation. I think there is literature (can’t remember where) that indicates that lecturers do change their position of lecturing a course depending on its department or discipline (or something like that) – I think that was in the Lindblom-Ylanne et al paper.
Well, I think John is pretty much sold on the idea of the disciplines – so he decided to give me some ammunition to combat James opposition by recommended that I speak to a Yann Lebeau who worked with him in the SOMUL project to help provide me with literature concerning the choice of how subjects are placed into which kind of discipline (i.e. soft or hard etc).
Well … with respect to the logistic regression … John wasn’t too certain what is the best road to take since he haven’t done logistic regressions in quite awhile. He suggested for those questions that have a small number of answers with ‘not sure’ to treat it as missing data (such as for the coverage and the delivery questions) and then I can treat the other 4 options as ordinal and hence can use ordinal logistic regressions. He, however, cautions I must check and see what the assumptions of logistic regressions are upheld and even if it isn’t and I go ahead and use it to remember to include in my discussion that I am treating this data as being normally distributed etc. when it is know it is and the results may not be quite correct.
Further, he suggested that I look up some non-parametric tests to check and see if I can use this for testing my data – he mentioned some kind of non-parametric one-way ANOVA which I can’t remember. Oh I’ve found it on the net – its called the Kruskal-Wallis One Way ANOVA . However, the problem with this is that it can only deal with factor I believe … wait let me check my facts … oh no that is not true since you can have a two way Kruskal-Wallis, well at least according to that website. However, I think it becomes more complicated than that – John did suggest using multivariate analysis … but that will assume that the scales were continuous – now John said for him the ATI they used (Intentions and Beliefs), that they were using underlying continuous scale and hence the values were continuous. I’m not sure if I am up to that point to believe it is continuous so will stick to it being ordinal at this point. In that case the ordinal logistic regression (well at least for the LP part)should be best … I can’t remember – but I have a feeling that the logistic regressions are non-parametric test since it uses chi-square values – if I remember carefully. Well, it does use Chi-square – so it is indeed non-parametric in nature and there is no need to uphold the assumptions of normality etc. But one website indicates that I should have at least 50 cases for each independent variable – sheesh – I don’t have that – so not sure what I am going to do in that case. However, there is a suggestion to use discriminant analysis instead but I think in that case I will lose the ordinality – plus need to be normally distributed and equal variances etc. Anyway, will have to examine normality and if it upholds might go with discriminant analysis because according to the website it is more powerful – but I’m more comfortable with logistic regression … we’ll see what I use.
Anyway, John also suggested in the case of ATI where persons left out about two or three questions we can assume it is not sure (but in order to this – you must do a missing value analysis and then decide where the cut off point is!). John also suggested that I can combine the values for Intentions and Beliefs and do a two part multivariate analysis (or was it called a double multivariate analysis) – since he said essentially they were the same thing (well got to base this on literature since he was basing his talk on that Norton et al, didn’t find any difference between them).
Further, in particular with respect to the questions on the delivery simplex algorithm etc. – John suggested (since I have the ‘not sure’ and the ‘not taught’ options and which might well be answered alot and cannot be treated as missing values) to combine the values and try to get an ICT variable – i.e. the amount that ICT is used in different disciplines.
Updated: Tuesday, 21 June 2005 3:46 PM BST
Higher Education Research Blog 