6

Part of my thesis tries to investigate the extent to which students "test Score" on the topic of electrical circuits in physic can be predicted from how they find the topic to be "interesting" ,Their "enjoyment" of the topic, and their perception of how "difficult" the topic is. I intend to use linear multiple regression analysis to do this, where "test Score" will be my independent variable and the predictors will be the "Student's enjoyment", "Students perception of difficulty" and how "interesting" student find the topic. I think student's test score is a continues variable. However the suspected predictors were obtained thus:

1. Working with electric circuits in physics is not interesting

Strongly Agree[ ] Agree[ ] Undecided[ ] Disagree[ ] Strongly Disagree[ ]

2. Working with electric circuits in physics is not difficult

Strongly Agree[ ] Agree[ ] Undecided[ ] Disagree[ ] Strongly Disagree[ ]

3. I do not enjoy practical work involving electrical circuits in physics

Strongly Agree[ ] Agree[ ] Undecided[ ] Disagree[ ] Strongly Disagree[ ]

the options are given these values:

Strongly Agree ---- 1
Agree ------------- 2
Undecided --------- 3
Disagree ---------- 4
Strongly Disagree - 5

Now since this is an ordinal scale, is it proper to use linear multiple regression analysis ?

Glen_b
  • 257,508
  • 32
  • 553
  • 939
Man Hunter
  • 61
  • 1
  • 3
  • 4
    No, ordinal variables cannot be predictors "as is". You should make a decision first: 1) either consider them scale (interval, but not necessarily equi-interval) 2) or consider them categorical. In the latter case recode them to dummies (as if nominal). When interpreting regression results, keep in mind the constraint that these are ordered dummies. – ttnphns May 13 '13 at 04:52
  • 3
    I disagree slightly with ttnphns. Ordinal variables may be predictors as delivered if when you examine relationships those relationships look linear. (Call them interval in the latter case if you will, but that's terminology in the mind of the analyst and not an incantation that affects the success of the analysis.) But on the whole, it's more likely that coding them as indicators (a.k.a. dummies) is the best way to proceed. – Nick Cox May 13 '13 at 06:43
  • 2
    I agree with @NickCox, it just depends if you decide the scale can be considered linear. If it can be, regression interpretation may be simpler. However if your purpose is purely test score prediction it is probably best to use dummy variables anyway – kirk May 13 '13 at 07:13
  • @Nick, where is your disagreement? To be able to observe linear relationship you must first _assume_ the variables are interval. As long as you call the variables ordinal you can only claim whether the relationship is monotonic or not. – ttnphns May 13 '13 at 08:04
  • 2
    The disagreement lies in your apparent faith in the power of terminology, i.e. that if you _call_ a predictor interval that makes the analysis justifiable. You are free to regard that as minor or unimportant. – Nick Cox May 13 '13 at 08:13
  • 1
    I believe the answer is given [here](http://stats.stackexchange.com/questions/77796/coding-for-an-ordered-covariate/77827#77827) and [here](http://stats.stackexchange.com/questions/33413/continuous-dependent-variable-with-ordinal-independent-variable). They refer to [this paper](https://epub.ub.uni-muenchen.de/2100/1/tr015.pdf) and R package [ordPens](https://cran.r-project.org/web/packages/ordPens/ordPens.pdf#page.10). – Ilya V. Schurov Jan 14 '16 at 10:50

0 Answers0