I am working on a research project to identify the relationship between sexual esteem and sadistic behavior. Both of my independent and dependent variable have been measured using standard questionnaires used in the field of research
* IV Sexual Esteem (5 point Likert scale, each item scored from 0-4 with a max scale score of 40): Snell and Papini, 1989
* DV Sadistic Behavior (1/0 dichotomous scale): Freund, Stein, & Chan, 1982
The problem is my DW is violating regression assumptions preventing me from running linear regression on my data. Since I am simply looking at correlation, most of the non-parametric tests does not suit my use case.
An alternative that works for me is to dichotomize my IV into LOW and HIGH categories. The split was done by categorizing scores > 20 as HIGH and scores < 20 as LOW. This allowed me to run meaningful non-parametric tests (such as Mann-Whitney U Test) which showed positive results.
However, dichotomizing seems to be discouraged in general, so my questions are:
1. Considering the situation I am in, is this something that I can explore further?
2. Are there any research that supports similar approaches?
You just need to run a binomial logistic regression. It assumes that your your dependent variable is measured on a dichotomous scale.
DICHOTMIZED LIKERT SCALE DATA. no such a thing. When respondents answer your questions, they answer according to the scale (1,2,3,4,5). if you take their responses and change it to (0,1) and says this is what they responded---which they did not---that would be faking data.
BINARY DEPENDENT VARIABLE. use logistic regression. https://en.wikipedia.org/wiki/Logistic_regression
I'm not sure that is faking data but it is certainly throwing information away. Don't do it. As suggested by Artyom Golossenko it appears logistic regression would solve your problem.. I 've attached a set of notes that should be helpful. Best, D. Booth
As folks say, there are few situations where you should dichotomize. See concern #3 in https://www.frontiersin.org/articles/10.3389/feduc.2019.00147/full . It describes this (and the author's mom says its a great paper).
For the actual model, I assume that you just want to make sure the model is has a monotonic relationship between these.
- Badges
- Science topic
- Similar topics
- Psychology
- Psychometrics