0
$\begingroup$

I have 8 predictor variables and 1 criterion variable, all variables are measured by 9 different constructs, some are 0-4 Likert scales, some are 1-5 Likert scales, can I use their mean scores and run a multiple regression?

$\endgroup$
8
  • 1
    $\begingroup$ Can you say more about your situation, @Shirley? Do you mean that you have 9 different Likert items, where some are numbered 0-4 & others 0-5? Or do you have multiple items (forming a scale) per construct, and you want to average over all the the items w/i a construct to get a single mean for each variable? Or something else? In formulating your question, it might help to read this blog post: how to ask a statistics question. Also, if your criterion is ordinal, you might consider ordinal logistic regression. $\endgroup$ Apr 26 '13 at 16:51
  • $\begingroup$ Thanks for prompt response and advice. Here is the situation: $\endgroup$
    – Shirley
    Apr 26 '13 at 17:12
  • $\begingroup$ I have 8 predictor variables using 8 different questionnaires. These 8 variables are stress level, self esteem, peer relationship, family relationship, neuroticism, boredom proneness, depression, social anxiety (So there are 8 different questionnaires). However, this 8 questionnaires have its each form of scales. $\endgroup$
    – Shirley
    Apr 26 '13 at 17:19
  • $\begingroup$ For example, Stress questionnaire is using 0-4 scale (0 = strongly disagree, 4 = strong agree); Self esteem is using 0-3 scale (0 = strongly disagree, 3 = strong agree); Peer relationship is using 1-4 scale (1 = strongly disagree, 4 = strong agree) and etc. My criterion variable is Online gaming addiction using 1-4 scale (1 = strongly disagree, 4 = strong agree) $\endgroup$
    – Shirley
    Apr 26 '13 at 17:19
  • $\begingroup$ So what I'm doing now is that I calculate mean score for these 8 predictor and 1 criterion variable (in other word, I have one single mean for each variable, so in total I have 8 predictor mean scores and 1 criterion mean score now). $\endgroup$
    – Shirley
    Apr 26 '13 at 17:21
3
$\begingroup$

@Shirley, using the mean scores for each of the 8 questionnaires, even though they have different response anchors (i.e., 0-4 or 1-7 or 1-4), is definitely acceptable in multiple regression. In fact, it is very rare that you'll end up having measures/questionnaires that use the same set of response anchors.

$\endgroup$
9
  • 1
    $\begingroup$ Thanks for answering my question! As I read from some stats articles, some researchers center or standardize the data. So I'm just confused.. $\endgroup$
    – Shirley
    Apr 26 '13 at 18:44
  • $\begingroup$ On top of the 8 predictors variables, I'm also measuring gender difference on the online gaming addiction (my criterion variable). So obviously gender has 2 levels (Males and Females). Since it is nominal data, I'm thinking to use independent t test to test out the gender difference on the criterion variable. Wondering for this particular test, do I still use the mean or centered mean? $\endgroup$
    – Shirley
    Apr 26 '13 at 18:47
  • $\begingroup$ One last question...Just to confirm, it is fine to use the mean even the response anchors involves 0, correct? $\endgroup$
    – Shirley
    Apr 27 '13 at 2:35
  • $\begingroup$ Hi Shirley, I think each of these questions could be separated as they touch on different regression related methodological issues. But just to give you an idea: $\endgroup$
    – PsycStats
    Apr 29 '13 at 14:47
  • $\begingroup$ Centering data is useful if you plan on running a moderation or mediated moderation analysis. For simple multiple regression there is really no need to standardize and center your data (as far as my knowledge of regression, but you may want to check with others). $\endgroup$
    – PsycStats
    Apr 29 '13 at 14:59

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.