Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Ologit for categorical variable cut off point

    Hi stata list


    I am using ordinal regression to check if the cut off points are sufficiently distinct. From reading the stata manual and previous questions on here, the cut-off point is the probability for one category to fall into the other and if the confidence intervals for two cut points overlap, then I might want to consider combining these groups if the N is too small for some of the categories.

    These are the results I have:

    Code:
    Iteration 0:   log likelihood = -6519.7348  
    Iteration 1:   log likelihood = -6519.7348  
    
    Ordered logistic regression                     Number of obs     =      6,335
                                                    LR chi2(0)        =       0.00
                                                    Prob > chi2       =          .
    Log likelihood = -6519.7348                     Pseudo R2         =     0.0000
    
    ------------------------------------------------------------------------------
    DMdailypurch |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
    -------------+----------------------------------------------------------------
           /cut1 |     .35702   .0255294                      .3069834    .4070566
           /cut2 |   .3726839   .0255655                      .3225765    .4227913
           /cut3 |   1.495419   .0324855                      1.431749     1.55909
           /cut4 |   1.592298   .0335213                      1.526598    1.657999
    ------------------------------------------------------------------------------
    
    . ta DMdailypurch
    
      RECODE of |
       q7201_02 |
         (Marr: |
      Final say |
     on mak. hh |
     purch. for |
          daily |
         needs) |      Freq.     Percent        Cum.
    ------------+-----------------------------------
              1 |      3,727       58.83       58.83
              2 |         24        0.38       59.21
              3 |      1,424       22.48       81.69
              4 |         89        1.40       83.09
              5 |      1,071       16.91      100.00
    ------------+-----------------------------------
          Total |      6,335      100.00
    If someone could kindly explain to me in laymen terms it it makes sense to collapse the categories with small N’s.

  • #2
    If we aggregate the categories 2 and 3, we'd still have less than 2% of the observations. In short, that may not help much in this specific case.

    This being said, the decision to aggregate categories should ideally have a sound rationale, i.e., a "reasonable" decision, not exactly from the statistical point of view, but mostly in terms of study design. Then, considering it went out as a reasonable decision (in terms of rationale), the large CIs may bite hard, this time statistically speaking.
    Best regards,

    Marcos

    Comment


    • #3
      Thank you for Marcos!

      -Sherine

      Comment

      Working...
      X