Tell me more ×
Cross Validated is a question and answer site for statisticians, data analysts, data miners and data visualization experts. It's 100% free, no registration required.
up vote 1 down vote favorite
1

I want to analyze some data of a questionnaire on presence (I deal with experiments on virtual reality). The questionnaire was provided twice to particpants performing the experiment, after each of the two provided conditions. The 6 questions of the questionnaire were evaluated by participants on a 7-points Likert scale.

Following what I found in bibliography, I have to analyze the data in the following way in order to find if the differences between the two conditions are significant: I have to count the number of answers that have a score of 6 or 7, then I will have for each condition a variable equal to the mean of the count of 6 and 7 scores among all the 6 questions. Afterwards, I have to treat those two variables as binomially distributed for a logistic regression on group.

My problem is that I don´t know how to perform the logistic regression. I tried to study but I do not understood how to apply it to my case. In addition I have also problems in undestanding how to perform it in R.

Do you also have an example in R? Let's say that the two variables are (meand +-std): 

Count_condition1 = 1 +- 1.7 and Count_condition1 = 2 +- 2.0
share|improve this question
1  
I would suggest checking out the UCLA tutorial on logistic regression in R to see how to do the software code. For further R resources check out this meta thread. You could learn alot for the interpretation and other aspects of logistic regression by perusing the other questions with the logistic tag on this site. – Andy W Sep 12 '11 at 1:59
Thanks a lot. I actually already had a look to the topics in the forum and also the material over internet. The thing is that so far I have not understood yet how to apply logistic regression to my case!...I don´t want to be wrong, I am very unsure so far, therefore I would like a small help to understand how you guys would perform this analysis. Expecially in R. Ant further suggestion?...maybe an explanatory example? Thanks in advance – Luca Sep 12 '11 at 6:23

2 Answers

up vote 1 down vote
  • Under most circumstances I would calculate the scale score as the mean of the items. This is generally a more desirable coding because you don't lose information. You could then just use a t-test to assess the effect of condition.
  • Even if you binary code each item, and calculate the mean or sum of items, you are still left with a variable with 7 different values (i.e., 0,1,2,3,4,5,6 if you used the sum). If you were willing to accept an approximation, you could consider still using a t-test in this case. However, you couldn't use standard binary logistic regression. Perhaps, generalised estimating equations (GEE) might be suitable for predicting repeated measures of proportions. R packages include gee and geepack. Here are some GEE resources that I prepared.

UPDATE

After reading the paper mentioned "Using Presence Questionnaires in Reality", I still think that the scale should be the sum or mean of the six items coded 1 to 7, and that a t-test is the most straightforward tool for group comparison of means.

share|improve this answer
Hi Jeromy, thanks a lot for your answer. Regarding the first point, it is exactly how I am not supposed to do. Instead I have to follow the method I explained, as a paper found in bibliography for this kind of study suggests. – Luca Sep 12 '11 at 6:19
@user4701 Why are you not supposed to follow the 7-point-mean,t-test approach? Could you cite the paper? – Jeromy Anglim Sep 12 '11 at 7:05
yes dear Jeromy (yaou are always kind in answering to all my posts on this forum :-) the paper is about measuring presence in virtual reality context. The paper I refer to is "Using Presence Questionnaires in Reality. (Usoh et al., 2000, Presence, volume 9, n. 5)"...I am using the SUS questionnaire there explained. In section "Results" it is explained why it is not correct to use the mean approach and the t-test....hope you can give me some suggestions about how performing that analysis in R...still I have not figured out yet properly ;-( – Luca Sep 12 '11 at 8:18
up vote 0 down vote accepted

At the end I solved the problem with a simple chi-square, since I had just to compare two variables whose possible values are 0 or 1.

Still I have not understood the need of having a logistic regression to be honest....

share|improve this answer
Unfortunately I believe some things have gotten lost in translation, and so it has been difficult for anyone on the site to more pertinent advice. Lets say you wanted to model more outcomes besides 6 or 7, had other covariates you thought might affect the survey answers besides the experimental condition, or wanted to compute the effect size of one condition versus another could all be possible reasons to use logistic regression. The Chi Square test has nothing (directly) to do with possible values of 0 or 1. – Andy W Sep 15 '11 at 19:16

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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