What's a logistic regression?


in college we learned multiple regression techniques (different models: linear, log-lin, lin-log, exponential, ...) - including all the assumptions of the error terms and how to test them.

Now I've heard of logistic regression.

Is this a lot different from multiple regression techniques?

Are there other (frequently-used) regression techniques other than those I mentioned above?

thank you
There are many, many types of regression. Logistic regression is just a regression in which your outcome variable is binary. For instance, if you want to build a model that predicts if somebody will pass or fail a class. There is a multitude of information about logistic regression online and in most standard regression books.

thanks a lot for the replies to my two threads, Dave!

It's exam time for me at the moment (and I dont have stats classes anymore..its more of a hobby) so I cant read up too much about logistic regression and how it works, right now. I'm still too curious about it not to ask, though :).

I was hoping to get an overview of the regression techniques (other than the OLS multiple regression analysis, I already know) that exist and which are applied in the real world (for example in business/marketing).

From what I've gathered there are these:

- the plain..normal..OLS multiple regression (the one which I already know)..either linear or as a log-lin or a lin-log or a log-log or a..... model

- the logistic regression..which is very similar to the OLS as you mentioned (Im not completely sure how it works, but I think I have a rough idea)

- the robust regression - which seems to not be put to use a lot in practice as it seems to be 'unpopular' (?)

- multivariate regression - which is the same as the OLS multiple regression just that multiple variables are changed at the same time (?)

- regression with time series data..which isn't really different from regressions with non-time data

- probably multiple combinations of the above techniques

Would you say these regression techniques are the ones that are used in most cases in a business/marketing setting? Or are there other regression techniques I left out that are used quite frequently?

thanks again!!
Analytics I did a quick read of Robust regression and the reason it's not popular is that standard Robust regression methods aren't very good. The MM method is superior however.

I doubt you're doing multivariate regression (which is a single predictor and multiple responses) and instead talking about multiple regression which is several predictors and one response. Multivariate typically requires a good deal of linear algebra and (maybe vector?) calculus to understand all of the theorems. It's usually an advanced undergraduate or grad school course in pure statistics.

Another kind of regression taught in usually a second semester course with a name something like "Generalized Linear Models" you learn about nominal and ordinal regression. These techniques are an extension of logistic regression. What if you have several non numeric outcomes with no way to order them? (This is nominal.) In the ordinal case, there is some order to them, but the categories are not numerical.
Hi Ironman,

I know the difference between multiple and multivariate regression (though I had to ask about it before :)).

Im not doing any statistics homework actually, but am looking to learn some advanced statistics myself, because I think it's fun and b/c it'll be good to know for my future job..Im looking to work either in web analytics or in business intelligence (most likely in web analytics where I have some experience and know the right people already, so having a business degree wont be a problem).

In web analytics multivariate regression is something that is used to test lots of diff. versions of a website for their effectiveness (as testing each variable at once would take too long which would make the test impossible in some scenarios). Actually it's not necessary to know how a multivariate regression works to do that (most web analysts are far from understanding it), but I was thinking if Im trying to learn some advanced stats that might something good to know about.

But then again, if its too complicated it might be better to only read up on the stuff I need to know without trying to study in depth (what I need to know to determine the sample size).

Any idea if multivariate regressions are used in business intelligence? or is it one of those not-used-because-too-difficult-to-explain-techniques (or too complicated to be effective in many real world examples)?


EDIT: Whats a MM regression?
Last edited: