How to modify continuous exposure variables in logistic regression?

[Mangji Zo]

Hi everyone,

I have a bit of confusion with how to modify the results from my binary logistic regression model for my continuous variables. I have attached part of my model to illustrate. The variables Ef_variablecat (kcal/ae/day) and hh_assetscat (Ugandan shillings) are both continuous.
I am currently interpreting them as for one unit decrease in kcal/ae/day, there is decrease in zwhwhocat by a factor of 0.25? I would like really some advice for the best way to interpret these variables.

Also, if I want to modify the units of Ef_variablecat to 1000 kcal/ae/day, do I need to modify the original variable (say divide by 1000) and then put in the regression model or can I directly apply changes when running the regression?

Apologies for asking such simple questions, this is my first time running a model like this. Thank you for any advice and help!

logistic zwhwhocat Ef_variablecat hh_assetscat

Logistic regression                             Number of obs     =        596
                                                LR chi2(2)        =       1.28
                                                Prob > chi2       =     0.5276
Log likelihood = -115.30864                     Pseudo R2         =     0.0055

--------------------------------------------------------------------------------
     zwhwhocat | Odds Ratio   Std. Err.      z    P>|z|     [95% Conf. Interval]
---------------+----------------------------------------------------------------
Ef_variablecat |   .7505161   .2364219    -0.91   0.362     .4047818    1.391551
  hh_assetscat |   .9645379   .1124659    -0.31   0.757     .7674825    1.212188
         _cons |    .077683   .0352254    -5.63   0.000      .031941    .1889312
--------------------------------------------------------------------------------
Note: _cons estimates baseline odds.

[Clyde Schechter]

Your interpretation of the odds ratio for Ef_variablecat is not correct. The output you are looking at is an odds ratio, not a probability ratio. What is true of your model is that a unit difference in Ef_variablecat is associated with a 25% lower odds of zhwwhocat. But that will not correspond to a 25% lower probability of zhwwhocat. (They will approximately correspond if the starting probability of zhwwhocat is < 0.1.)

To change the units of Ef_variablecat, change Ef_variablecat itself (divide it by 1000). Because the logistic model is non-linear, multiplying the odds ratios by 1,000 will not produce equivalent results (the way it would if applied to linear regression coefficients.)

[Mangji Zo]

Thank you Clyde for the corrections! I learned a lot from your post.