Data transformation for bacterial count data

[Rayhan Islam]

Hello Stata Altruistic,

In my dataset I have faced some problems with bacterial count data.

Here TBX= Total E.coli count, ESBL= Total beta lactamage e.coli count,
Now i want to work with ESBL proportion (ESBLProp=ESBL/TBX) for further analysis but for 0/0 values corresponding ESBLProp cell remain blank. So in analysis it consider as missing value. I know 0/0= Infinity. To remove these problem i have taken some transformation:

gen logTBX= log10(1+TBX)
gen logESBL=log10(1+ESBL)
gen ESBLProp=ESBL/TBX
gen logESBLProp= log10(ESBLProp)
gen logESBLProp1= log10(1+ESBLProp)
gen logESBLProp2= logESBL/logTBX
gen tESBLProp= asin(sqrt( ESBLProp ))
gen tESBLProp1= asin(sqrt( ESBL/TBX ))

My question is
1. Which transformation is used in these situations? I searched and got above mentioned transformation but need reliable transformation method.
2. In addition it very much logical if you don't have TBX then you don't have ESBL count as well, what will be steps for further analysis, treat these as missing value or not?
3. I'm pretty sure I can use GLM, ZIP for ESBL, what will be better choice for dependent variable ESBL or logESBL?

[Maarten Buis]

None of the above. I would consider a poisson of ESBL with TBX as an offset().

[Rayhan Islam]

Thanks for your prompt reply. Why none of the above conversion is right choice. Would you please explain.
For bacterial count data people use log transformation every now and then.

[Nick Cox]

See the concurrent thread https://www.statalist.org/forums/for...predict-values