what is the appropriate statistical analysis to show correlation between number of hours spent for online classes and its perceived effect on students’ mental health? The number or hours was one multiple choice question with 4 choices and the perceived effect on students’ mental health is a likert scale of 5 intervals indicating likelihood (never, rarely, sometimes, often, always) with 10 different questions. I see people recommending Pearson or spearman and i am thoroughly confused since i have ten questions. how do i condense the ten questions into one variable?
furthermore, i was initially planning to see the effect of online learning to students perceived mental health, but did that mean using mean and standard deviation. is that correct analysis for the hypothesis “online learning does not have an effect to students’ mental health”?
First you need to look at the literature on building scales from Likert questions. Hopefully I am using appropriate terminology. I see @Holger Steinmetz is reading this question. He and @David L Morgan are both well qualified to discuss this issue with you. You might discuss the issue with them. Both are very active on Reseachgate.. I would suggest that the variables seemed based on Likert questions (ordinal) that you look at ordinal logistic regression with mental health as your DV based on your variables and what I think is your Research question: planning to see the effect of online learning to students perceived mental health, Once you decided this issue and say googled Ordinal logistic regression you can decide if these comments are appropriate to your goals. At that point I would suggest returning here for a more detailed discussion of methods. Best wishes, David Booth
Hello Ja Su,
A couple of general observations about Likert-type scales (though do have a look at the resources given below):
1. The original conceptualization of a scale was that the researcher would combine responses across a set of items (usually by summing the 1-5, or whatever arbitrary values are given to a single response) that are unidimensional--that is, they all pertain to the same target object, concept, attribute, characteristic, person, or idea. If sufficient items were included, then: (a) the resultant score would be more reliable than any single response: and (b) the resultant score would tend to behave like a continuous variable (technically, it's still ordinal, but a lot of practitioners treat it as if interval strength).
2. Item response theory models allow the aggregation of such items (again, dimensionality is a consideration) in a way that yields interval strength scores for the resultant scale.
3. Factor analysis would offer yet another way to combine scores over a set of items.
4. The tendency of researchers to use single items which use a Likert-type response scale as a variable is not best practice.
5. To answer your apparent question, good old Spearman correlation would be satisfactory.
Now, for the resources:
David Morgan has posted a lot of information on how to handle a set of Likert-type response items here on ResearchGate. You could also look at Rensis Likert's original monograph (see this link: https://legacy.voteview.com/pdf/Likert_1932.pdf). A more recent, and somewhat simplistic view of Likert-type scale item sets may be found here: hhttps://www.scribbr.com/methodology/likert-scale/
Good luck with your work.
You could run an Item Analysis (for internal consistency) on the 10 items. This is a correlations analysis with the Cronbach alpha to select the reliable items for Scale (and or SubScales).
(See, e.g.:
https://blog.minitab.com/blog/meredith-griffith/item-analysis-with-cronbachs-alpha-for-reliable-surveys
or
https://mran.microsoft.com/snapshot/2017-04-11/web/packages/sjPlot/vignettes/sjtitemanalysis.html)
Thereafter, you can calculate correlations between the Scale(s) and the hours spent online. Check the normality of the distributions of the hours spent and the Scale(s). If fairly normal (look at the histograms), just trust the Pearson correlations.
- Badges
- Science topic