Hello Bhushan,

What is the dependent variable, R? How is it quantified?

Are your "factors" actually estimates of scores on latent variables (as in factor analysis), or just topics/specific attributes over which you've posed one or more questions/stimuli to your respondents?

How many items did you average to yield a score for each variable (A/B/C/D)? Did you check each variable set for unidimensionality (so as to affirm that it makes sense to combine these responses via an arithmetic mean)?

Finally, could you explain what you mean by "...if these averages are statistically significant" (for what purpose?) and "...then rank these factors" (based on what outcome)? Perhaps more basic, what specific research question are you trying to answer?

Given this additional information, I'd wager that you'll likely receive a more focused recommendation on how to proceed.

Good luck with your work.

Just to add a comment after you form scales suggested as done by David L Morgan (on RG) you might want to try ordinal logistic regression if appropriate for your research questions. Google search for this topic there are many references. Best wishes David Booth

Like David Morse, I would need to know more about R (the response variable) to provide an answer. I would also need to know more about A, B, C, and D. Are the subscale scores, or are they scores from four different Likert scales? Either way, I would suggest checking to ensure they are unidimensional with a Cronbach's alpha test. If they are subscales, you may want to consider using factor scores instead of averaging the item scores. What type of regression analysis you use would depend upon the characteristics of the response variable.

Hello again Bhushan,

Clearly, you did have something in mind when you posed your query, but it appears that this hasn't yet been very clearly communicated. Hence, all those responding thus far are asking for more information!

Are you trying to rank relative judgments about some specific features of some target? (For example: What's most important to peoples' perceptions of some product's desirability...cost, reliability, quality, status, ease of repair/replacement?)

Are you trying to see whether participant characteristics relate to the ratings they would assign some target? (For example: Do men and women rate the target comparably?)

Are you trying to determine if relative judgments about some specific features are statistically distinguishable? (For example: Do people rate friendliness of staff and vendor location equally among reasons to shop at a certain store, or is one rated significantly higher than another?).

And so on...

If you could help us understand what you're after, and what you're working with (the "factors" you mention; perhaps sample items for each), I would imagine you would be more likely to get a focused recommendation.

Good luck with your work.

Hello Prof.Morse,

Yes, I am trying to rank relative judgements about reasons why people download an app. And I did a likert scale survey for it. But I wanted to know there is any statistical test which I need to do before submitting the results.

Hello again Bhushan,

So, you have a set of ratings for each of a set of participants, and you'd like to compare the ratings for equality (vs. the alternative hypothesis that they are not equal).

Given the nature of the scores, I'd suggest you have a look at the Friedman two-way anova by ranks. This is suitable for non-independent data points (which the repeated measures nature of your list of reasons happens to match).

If you simply wish to compare pairs of reasons, two at a time, the Wilcoxon signed ranks sum test _or_ the Friedman test could be used.

Good luck with your work.

Hello Prof. Morse,

I am comparing each of the ratings with one standard rating. And then my hypothessi is that each of the rating is better than that standard rating. Sowhich statistical test to use since my sample size is more than 80. However for some ratings the mean is already lower than the standard rating. Should I drop these lower mean ratings from my further statistical tests?

In summary,

I have 80 participants

Participants gave a likert scale based likelihood rating for variable R for each Factor A, B , C, D and I am compring it with standard factor S which also has a likert scale based likelihood rating

R is dependent variable, factors ABCDS are independent conditions or variables

So whichh statistical test to use to check if mean rating is hogher are higher for each factor than S

Hello again Bushan,

The DV is the rating given, on a Likert-type response scale.

The IV is the "factor" or feature being rated. It has five levels (A, B, C, D, and the "standard" or "baseline" level, S).

However, since you wish to compare each level/feature (A-D) to S (implying 4 pairwise comparisons), it sounds to me as if the Wilcoxon signed_rank_sum test would be a viable option (as mentioned in an earlier post). I'd avoid using means as the summary statistic of interest, given the ordinal nature of the scale, unless you had somehow contrived the response scale to represent equal intervals of "likelihood."

Good luck with your work.

I was thinking to use t test since sample size is greater than zero and central limit theorem. Also I designed scales during survey with numbers.

However my null hypothesis is that :

Mean of R > Mean of any of the factors

and Alternative is Mean of the factor > Mean of R

What changes do I have to make in normal t test ?

also I have decided not to rank the factors and just check which factor is better than R

Hello Bhushan,

I'd avoid t-test / average scores and distributional assumptions altogether, and stick with something that honors the ordinal nature of the scale.

We're now at four postings, and a reader of this thread still can not deduce exactly what you have, or how it is quantified. This is one of the reasons that consulting with statisticians prior to collecting data is by far the easiest way to avoid issues later on down the road.

I'm signing off this one, but wish you good luck with completion of your work.

bjr Monsieur, votre question ne me paraît pas claire. Quelle est la qualification de ta variable expliquée.

En ces temps ci je suis trop pris pour te communiquer sur resaerchgate. Joins moi par mon mail : [email protected]