Appropriate Regression model to measure how attributes influence decision making

[Rodrigo Cepeda]

Hello, i need help to identify wich model should i use to answer the following question, "attributes that influence the decision making of buying a vegan product"
My dependent variable is the choice to buy or not a product (binary variable, 1 if they buy, 0 if they dont) and my independent variables are 5 attributes (factor variables, im using a scale from 1 to 5, 1 being the highest rank). i was thinking of a logit model but im not sure how should i work with this kind of data.
i used a test data and used this command "logit id gpa i.atb1 i.atb2" and got the following output:

If anyone is reading this, Thank You very much.

[Phil Bromiley]

You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output (fixed spacing fonts help), and sample data using dataex. Also, screen shoots and attached files are not encouraged.

Logit or logistic are quite appropriate. They differ whether they're giving you betas or odds ratios parameters. I prefer logit with betas but that is a matter of taste.

You are close. Looking at this, the coefficients are versus the omitted category (value of 1). So, atb2 of 2 results in significantly less likelihood of purchase than atb2 of 1, etc. You can use test after the estimation to see if the parameters for 2, 3, and 4 differ significantly. You can also use margins after logit to look at differences in predicted probabilities. You want to do this because unlike a linear regression, you can't simply look at the logit betas and say something about magnitude of effects.

[Weiwen Ng]

Also note, I believe you had previously posted this on Reddit, and you got only mildly informative answers. Do note that the FAQ asks us to notify the community if you have cross-posted - this is to prevent double-answering the same question.

I'll answer here. We don't know your own data. Your above regression is from a stock dataset on the UCLA website. If the dependent variable was binary, this is appropriate.

You specified the two independent variables as categorical variables, so you get the difference versus the base level of each one (e.g. versus level 1 of atb1). This is appropriate as well. One poster on Reddit said that you can just treat your independent variables as continuous in some disciplines. I'd prefer the categorical specification. You can't go wrong by doing that.

If you have more specific questions, you need to ask them specifically, e.g. what is a logit unit or an odds ratio? If you know what these are but you can't quite say clearly how they translate into probability, then don't worry: few people can do so, but you can use Stata's margins command to show the changes in probability terms.

[Rodrigo Cepeda]

First of all, thanks for answering, it seems that i figured it out. currently im gathering the data and i hope that my model will work. will take note on the margins tip.
also i did not know the reddit was connected to this forum (seems pretty obvious now), sorry about that.

[Weiwen Ng]

First of all, thanks for answering, it seems that i figured it out. currently im gathering the data and i hope that my model will work. will take note on the margins tip.
also i did not know the reddit was connected to this forum (seems pretty obvious now), sorry about that.

Reddit is not specifically connected to this forum. Some of us may post there occasionally (and our user names may or may not be the same). Remember, per the FAQ, cross-posting is totally OK, but we do ask that you say so.