1

I am (trying) to conduct an ordinal logistic regression for the first time using the polr function in r. I seem to be encountering a problem with the order of my levels of my DV.

My DV should have four ordered levels: 'no problems', 'low risk', 'moderate risk', and 'problem'. I think that we have coded it in this order:

    mutate(pgsi_label = factor(t2_pgsi_cat, 
                               levels = c("non_prob",
                                          "low_risk",
                                          "mod_risk",
                                          "prob_gamb"), ordered=TRUE,  
                               labels = c("Non-problem gambling",
                                          "Low-risk gambling",
                                          "Moderate risk gambling",
                                          "Problem gambling"))) %>%

When I run the analyses, the intercepts seem to indicate to me that the categories are being entered in the wrong order:

# regression model
fit <- polr(t2_pgsi_cat ~ [covariates], Hess=TRUE, data=kv_longitudinal_state)

# output

Intercepts:
                   
low_risk|mod_risk  
mod_risk|non_prob  
non_prob|prob_gamb

I assume the intercepts should read:

non_prob|low_risk
low_risk|mod_risk
mod_risk|prob_gamb

Any ideas if this is an error and what might be the source + solution?

Thanks!

nicola
  • 21
  • 2

1 Answers1

1

We found the solution

Needed to first specify the ordered levels:

pgsi_cat_levels <- c("non_prob",
                     "low_risk",
                     "mod_risk",
                     "prob_gamb")

And then use these levels:

    mutate(pgsi_label = factor(t2_pgsi_cat, 
                               levels = pgsi_cat_levels, 
                               labels = c("Non-problem gambling",
                                          "Low-risk gambling",
                                          "Moderate risk gambling",
                                          "Problem gambling"))) %>%
nicola
  • 21
  • 2