0

I have data set of different school in a certain area to assess the cause of reason behind student's cause of failure in the English exam. Data set has several kind of variable like continuous, categorical, variable with percentage value 0 to 100 etc.

DV : Percentage of student pass in the English exam (0 to 100)

IV : - No of students in class (Discrete value)

     - Average age of the teacher (Discrete value)

     - Percentage of teacher has masters degree or higher (0 to 100)

     - Quartile rank of students economic condition (1-4, poor to rich)

     - Percentage of students get free meal at the school (0 to 100) etc.

I want to run a ordinary least square regression in the data set, is it possible to run this on the above mentioned variables or I need some modification before analysis.

I have already run the linear regression for the model and surprisingly got all are statistically significant. So my questions are 1. Is it correct? If not what are the flaws? 2. Does it necessary to use logistic regression by converting to category?

Md. Rayhanul Islam
  • 1
  • 4

1 Answers1

1

It's definitely possible. But whether or not it's a good idea is an entirely different question. The problem with the approach you have taken is that nothing is preventing your model from making predictions outside the valid range (0-100). You should probably use logistic regression, which constrains your predictions to be from 0 to 1.

Also, the situation you describe will probably benefit from causal inference methods such as propensity score matching as I can imagine you'll have a lot of problems with confounding in this study, and you'll want to be sure you are comparing apples to apples.

StatsStudent
  • 10,205
  • 4
  • 37
  • 68
  • In this case, Beta regression seems more appropriate as the dependent variable is a % ranging from 1 to 100 (why not 0?) instead of simply being a binary (say 1 vs 100) variable. – Nicolas K Jun 18 '17 at 20:10
  • No need to overly complicate matters with the introduction of Beta regression, when a simple scaling of the data to a proportion (essentially the same thing as a percentage) will suffice. – StatsStudent Jun 19 '17 at 00:17
  • @Umka I have corrected it. – Md. Rayhanul Islam Jun 19 '17 at 09:36
  • @StatsStudent thanks for your suggestion. The thing is I have to use ols as per my mentors recommendation on this data set. I think I have to convert the variable's value (0-100) to (0-1) and run regression. It will great help if you share any resource. – Md. Rayhanul Islam Jun 19 '17 at 09:43