0
$\begingroup$

Does anyone know what exact data cleaning steps one need to undertake in order to clean data for a logit regression (not a logistic regression)?

I have only time variables, meaning year and month, as my independent variables, and I am using R.

A logit regression is simply a normal linear regression where the DV have been transformed with the following formula:

logit(y) = ln(y/(1-y) for

An example:

3 of 12 people gets cured from taking a pill in period 3 -> ln(0.25/(1-0.25)

5 of 25 people gets cured taking a pill in period 5 -> ln(0.20/(1-0.20)

One can use the logit transformation if you have ratios and in many papers and books it is closely related to the logistic regression.

$\endgroup$
  • 1
    $\begingroup$ I don't know logit regression. Can you give more details?(especially because you say it is NOT logistic regression) $\endgroup$ – Momo Mar 9 '13 at 21:13
  • $\begingroup$ Sorry. Question changed, with better explanation. $\endgroup$ – Kasper Christensen Mar 9 '13 at 21:50
  • 2
    $\begingroup$ OK, I see. Although it won't help you directly with your problem, I don't think you should use this technique as the equal variance and mean and variance are independent assumptions are likely not met. An alternative might be to use a (quasi-)binomial model if you know the successes and trials or a Beta model if you have only proportions. $\endgroup$ – Momo Mar 9 '13 at 22:19
1
$\begingroup$

Data cleaning is data cleaning. It doesn't matter whether you are doing regression (any type) or something else, bad data will mess things up.

So... Are any of your DV incorrect and or impossible? Are your dates correctly coded?

etc.

But, if you have many time points, shouldn't you be looking at some sort of time-series model? Regression (any type) with time series data is fraught.

$\endgroup$
  • $\begingroup$ With regards to data cleaning I am pretty sure I cannot do anything else to my IV's than I already did. It comes down to my DV which is a logit transformation of the proportion of event within a given month. I am thinking that I simply just might remove anything that is more than two-three standard deviations away from the mean, but I am a bit unsure if that is simply just it...? Hmm, so it is not reasonable to look at "January" as a predictor of an event, or whether a given year had significant influence on my DV? $\endgroup$ – Kasper Christensen Mar 10 '13 at 15:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.