Logistic Regression

trinker

ggplot2orBust
#4
Used pretty much the same as typical regression except the outcome variable is binary (0/1). The results usually tell us a model fit and how the predictors decrease/increase the odds of the outcome occurring.
 

gianmarco

TS Contributor
#5
Hi!
Just to complement Trinker's useful summary: it happened to me to use LR in an article I wrote. I tried to summarize LR in layman's terms, citing further references.
The article is freely available here.

Hope it will help a bit
Regards
Gm
 
#7
Gianmarco

Interesting stuff...I can hear my old arch tutor grumbling about processualists :-}

Also, thanks for clarifying a few points - I recently posted a GLM type question to the Applied Stats thread, and I am trying to figure these things out in my head too....still not sure I have got it quite straight yet, but I will, hopefully, get there. I assume that if you can test for differences between data and model, you could also test for differences between subgroups in the data? (My problem has presence/absence and categorical data)

Cheers

Paul
 

rogojel

TS Contributor
#8
hi pnwright,
the logistic regression with categorical IVs will do something quite close to what you are interested in, if I understand correctly. It will create indicator variables for the categories you have, define one as a base and compare the outcome of each of the other categories to the outcome of the base variable you picked.

To make it concrete, imagine that you have 4 bait types, rabbit, chicken, lamb .. etc. When you do this type of analysis you will pick one bait type as the basis, say meat. The logistic regression model will give you an estimate of the ratio of odds of seeing an animal in the trap with meat as a bait and the odds of seeing an animal with chicken as a bait.. and so on for each type of bait. It does not compare each bait type to each but all of them to the one you picked as a base,

From the odds ratios you can go to an estimate of the probailities and get the probability of seeing an animal in the trap for each bait type.

regards
rogojel
 

rogojel

TS Contributor
#10
:) yepp, it was pnwright´s original example in the other thread, I just could not remember the other bait types. did consider rat though.
 
#11
Thanks folks....I coded my baits 1, 2,3 etc., then coded the other column 1 or 0 if the camera trap too a picture of the mammal.

I've followed the example here: http://www.ats.ucla.edu/stat/r/dae/logit.htm

I *think* the output from the Wald's test showed that bait type was significant, and that one bait was significantly different to two others were not....but that's if I'm reading this all right

Paul
PS Baits no more exciting or endangered than chicken, rabbit and pigeon!! No dolphins were harmed in this field survey.....
 

gianmarco

TS Contributor
#12
Gianmarco

Interesting stuff...I can hear my old arch tutor grumbling about processualists :-}
Hello,
thank you for your comment....
Yes, when people (i.e., archaeologists) read articles in which archaeology goes hand in hand with statistics, they start to think to "processualism" and to its limitations. Indeed, I believe that archaeology/anthropology has made a long way beyond processualism, but sometime numbers and statistics help to understand what is going on in data...Quite fortunately, my article (and its theoretical backdrop) leave much room to human agency, yet trying to set a more general stage in which human choices (cultural determined) can be put to work ;-)

Cheers,
Gm