1
$\begingroup$

In my research I have one concept consisting of six attributes. This whole concept forms my dependent variable. So basically I have six dependent variables measured on ordinal scale (five point Likert scale).

I have three independent variables, also ordinal (five point Likert scale). From here Logistic regression with ordinal variables I got that I can use ordinal regression method by converting my ordinal independent variables to categorical (because this is my main problem now, I don't understand which method to take since my independent vars are ordinal)

My research questions are to find positive influence of every independent variable on the whole dependent concept (or on every attribute of the concept).

  1. Is this approach achievable in SPSS?
  2. and would it work and produce valid results?
  3. "converting" ordinal independent vars to categorical is not "recoding"?
$\endgroup$
1
$\begingroup$

1) You can either use the Order Logit regression or the Order probit regression.

I do not know whether this approach works in SPSS, but here there is a nice code for the Order Logit Regression in R.

library(MASS)
m <- polr(independentvar ~ var1 + var2 + var3, data = ghost291data, Hess=TRUE)

2) You get the following output:

  1. A list of coefficients like for any regression
  2. Two intercepts which indicate the differences between the different ordinal datas. You will get n-1 intercepts for n categories of the independent variables.

3) The Algorithm from the MASS package does the recoding for you.

$\endgroup$
  • $\begingroup$ Thanks Ferdi, but Probit regression, also called a probit model, is used to model dichotomous or binary outcome variables. My dependent variable is not dichotomous or binary, it's ordinal (scale from 1 to 5). $\endgroup$ – ghost291 Sep 26 '16 at 17:58
  • $\begingroup$ @Ghost291: You are right probit or logit regression is for binary outcomes, but ORDINAL LOGIT or ORDINAL PROBIT is for ordinal outcomes. For a simple read with application in SPSS I suggest the following page: ats.ucla.edu/stat/spss/dae/ologit.htm $\endgroup$ – Ferdi Sep 27 '16 at 6:15
1
$\begingroup$

Ordinal independent variables should be treated as factors or, equivalently, converted to a set of dummy variables. Then any regression-related procedure can be used.

For the dependent variable, the PLUM procedure in SPSS Statistics handles this. Since some people consider Likert variables to be scale/continuous, you might want to compare the results with ordinary regression as well.

$\endgroup$
  • $\begingroup$ thanks JKP for the precise answer I was looking for. Now it is clear. When doing PLUM in SPSS I consider my independent ordinal vars as factors. that's it. $\endgroup$ – ghost291 Sep 27 '16 at 19:42
  • $\begingroup$ thanks a lot guys. but I have a doubt about how I coded my independent and dependent vars. So my value-labels in SPSS looks like 1="Very good", 2="Good", 3="Fair", 4="Poor", 5="Very poor". If I code in this manner dependent and independent vars is it ok to run PLUM ? $\endgroup$ – ghost291 Sep 27 '16 at 20:21
  • $\begingroup$ From the PLUM help..."The dependent variable is assumed to be ordinal and can be numeric or string. The ordering is determined by sorting the values of the dependent variable in ascending order. The lowest value defines the first category. Factor variables are assumed to be categorical. Covariate variables must be numeric." So no recoding of the dependent variable is needed. For the independent variables, you can just specify that they are factors. If you were to use ordinary regression, you would need to convert them as suggested above. $\endgroup$ – JKP Sep 28 '16 at 13:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.