Generalized linear Model setup and interpretation

[Rayhan Islam]

Dear Stata Altruistic,

I have data set like following one
where ESBLProp = ESBL/TBX

* Example generated by -dataex-. To install: ssc install dataex
clear
input double(TBX ESBL) float ESBLProp double(StudySite Status Season)
  3600000      0           0 1 1 1
  5000000      0           0 1 2 1
 25000000  46000      .00184 1 1 2
1.300e+08      0           0 2 2 1
 76000000  18000 .0002368421 2 1 1
4.700e+08 170000 .0003617021 2 1 2
  1700000      0           0 3 1 1
  1000000      0           0 3 2 1
3.400e+08 150000 .0004411765 3 1 2
 42000000 130000  .003095238 3 2 2
end
label values StudySite StudySite
label def StudySite 1 "Rural" 2 "Farm" 3 "Market", modify
label values Status Status
label def Status 1 "Case" 1 "Control", modify
label values Season Season
label def Season 1 "Dry" 2 "Wet", modify

Now I use glm with following code

glm ESBLProp i.StudySite i.Status i.Season, family(poisson) link(log) robust
glm ESBLProp i.StudySite i.Status i.Season, family(binomial) link(probit) robust

So I have three questions

1. Which code is perfect in context of above dataset (probably Poisson cause outcome variable is continuous rather than binary)?
2. Is it any problem that there is no quantitative variable as independent variable?
3. If first command is right what will be the interpretation of the outcome (like Poisson regression)?

Any other suggestions will appreciate.

Thanks
Rayhan

[Marcos Almeida]

If I understood right, the dependent variable is a proportion. You may search on the Forum posts as well as on the Web for the ‘most appropriate’ models under this scenario. In short, a glm with a binomial family and a logit link will work fine. Hopefully that helps.