1
$\begingroup$

I'm trying to see how level of scepticism impacts willingness to change diet. To measure sceptism I've used a 7 point likert scale. The study I'm basing my research on used a principal components analysis when analysing the Likert scale results to measure level of sceptism. I've got this far. But then the study uses a multinomial logistic regression to assess how skepticism effects willingness to change. Any idea on how to take the results of the PCA and put them into the MLA? (I'm using SPSS)

$\endgroup$
  • 1
    $\begingroup$ Why did they use multinomial logistic regression? What is your response variable? If it's just willing / not willing, then regular logistic regression is appropriate. What are you / they doing PCA on? If you have only 1 var, there is no point in doing PCA. $\endgroup$ – gung May 7 '15 at 19:41
  • $\begingroup$ the response variable is 'certainly', 'I'm already doing that' and 'No'. The PCA was conducted on 5 items on the likert scale. The study says 'the factor score of the first unrotated component was used as a measure of skepticism.' I don't have to follow the study, I'm just interested in measuring skepticism from the likert scale and then comparing it to willingness to change $\endgroup$ – Tim Rose May 7 '15 at 19:48
  • $\begingroup$ Do you also have 5 likert items, or only 1? Are you interested in all of those responses individually? If so, would it be reasonable / would you be willing to think of them as ordinally related (ie, already > certainly > no)? $\endgroup$ – gung May 7 '15 at 19:54
  • $\begingroup$ Yes, also 5 items. no I'm more interested in making a measure of skepticism in general, not each individual item. I'm beginner so I'm interested in doing what is going to be easiest for me, I'm not too sure what ordinally related means $\endgroup$ – Tim Rose May 7 '15 at 20:00
  • 1
    $\begingroup$ No, I mean for your response variable you can have a categorical variable w/ 3 nominal / unrelated categories, or 3 ordinally related categories. If you think the latter is reasonable, you can use ordinal logistic regression instead of multinomial LR. The former will be more powerful & informative. $\endgroup$ – gung May 7 '15 at 20:04
1
$\begingroup$

They obtain 5 explanatory variables. Then they apply PCA and get its first component score. Next, they run multinomial logit using that score as an explanatory variable, instead of the original 5 variables. PCA score is nothing more than a weighted average of the variables, except that the weights are chosen in a certain way.

I'm skeptical of this kind of studies. I bet you can get the same or similar result with a simple average of these 5 variables instead of PCA. In my opinion PCA is just a fancy method which doesn't add anything to this kind of data. It only makes it look more scientific.

$\endgroup$
  • $\begingroup$ Ok. Think I'll stick with this method and then discuss the limitations. What I'm struggling with is taking the factor scores and running them into the multinomial logic. I'm new to SPSS and honestly don't know how to make it work $\endgroup$ – Tim Rose May 7 '15 at 20:12
  • $\begingroup$ This forum will not help you with the code. People here tend to answer statistics questions, not coding $\endgroup$ – Aksakal May 7 '15 at 20:15
  • $\begingroup$ ok any idea of what forum could help me with coding? $\endgroup$ – Tim Rose May 7 '15 at 20:53
  • $\begingroup$ I can't speak for SPSS but SAS has both a technical support service for currently licensed users as well as a "community" support forum which permits Q&A regarding SAS procedures for any and all users, licensed or not. As big a product as SPSS is, the chances are high that something similar is available for SPSS users. $\endgroup$ – Mike Hunter Nov 8 '15 at 13:01
  • $\begingroup$ @TimRose I'd try StafkOverflow with SPSS tag. $\endgroup$ – kaqqao Jan 14 '16 at 11:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.