An Intuitive Introduction to the Multinomial Logit

[@unicornsandrainbowsandchic2336]

8 years later and you are still saving lives. Thank you, sir.

[@abhilashashibu7878]

The concept was very well narrated along with all useful commands. I am using this model for my master's dissertation and would like to acknowledge your guidance. Thank you so much!

[@rayskum1]

This is the most succinct explanation I have come across about these models. Thank you.

[@ottilliaanalytics3804]

At 26:30 , female BRR (.590754) should mean, compared to male, female seems to have LESS odds of preferring chocolate to strawberry.

[@kriti3011]

I had the same question. Thanks!

[@JanaeBonsu]

This video is a godsend. Thank you so much for this!

[@irenerizzoli7157]

Great video, very useful to understand the multinomial logit! Thanks

[@adityarazpokhrel7626]

Thank you sir. Greetings from Nepal.

[@davila1906]

Thank you so much for this, it has helped me a lot!

[@xenonmob]

Thank you, exactly what I needed!

[@kristincochran893]

This is excellent! Thanks so much.

[@AnonymousIguana]

26:30 "being female seems to have the odds of preferring chocolate to strawberry relative to male, but it's not significantly different from no effect at all". Are you suggesting that being female increases the odds of picking the chocolate flavour relative to being male? I think that it's the opposite here.

[@someguy1169]

At 16:11, can we also say that someone with a low ses, is 4.84 time more likely to like ice cream? Meaning if this data was regressed for *only* dummy variable ses=1, would that be the odds ratio?

[@felipedrada]

Good video! I personally missed a little bit of a review (regarding the "margin" and how to plot the results from the inference) using the alligator data. All in all, a nice wrapup.

[@beatrizvidovichi6175]

amazing video, THANK YOU!

[@garimadhir4777]

At 31.20 minutes - based on the chi2(4) = 10.55 and p value = 0.0321 - we can reject the Null (that they are all zero), right?

[@ronnyzhu2465]

This is great! Thank you so much

[@AnonymousIguana]

11:54 "that's the predicted probability...". Shouldn't it be a "odds" rather than "probability"?

[@thiagocanhoto5718]

Thanks so much for this explanation, but i have one doubt, one of my independent variables is too big and i dont need their marginal effects but i need them in the regression to get better results, is there anyway to use them just to better results on the other indepent variables? like when using the commnad absorb in the ols regression in stata. Thanks!

[@enricacroda444]

Very nice, thanks!

[@iizdiananyachieo7326]

I don't understand why at the 27th minute we say kids from middle aged homes have a lower odd of preferring chocolate to strawberry relative to kids from low income groups. Why is it not higher given the rrr is positive?

[@farooqahmed2234]

thanks for important efforts

[@jarudify]

Im confused at 35:37, u said the margins show the probability of choosing VANILLA ice cream, but isnt outcome(1) the flavor for chocolate? Thanks for the video btw

[@dougmckee673]

lala wonder You are absolutely right that this is confused in the video. I refer to outcome 1 as vanilla, but in the data itself 1=chocolate and 2=vanilla. Ugh. I'm going to have to rerecord at some point and fix that. I'll also try to fix the drab monotone voice--for the record I don't talk like this when I'm in front of a class! Thanks for letting me know about the problem.

[@BeyondtheClassroom]

Great video

[@paulagonzalezmartinez7281]

Hi Doug! Thanks so much for this explanation. What I am having is that when I compute the marginal effects for a multinomial logit some of the marginal effects change direction and significance. Do you know why this is happening and if is possible to correct it in any way? Does this indicate a problem in the module? Thank you!

[@sumonkyaw6909]

Hi .I have the question .Could we use SPSS software for multinomial logit model ? Please instruct me

[@clavink7948]

yes it's possible

[@kamalpreetrakhra8071]

I have a question. Is there a precedence of taking a random sample of one category of the dependent variable so as to have similar proportions to the second for a three category dependent variable. My category proportions are 0.77, 0.20, and 0.027. Is there any other way to model the three category dependent variable for these proportions.

[@keithhullenaar6487]

Hello Doug, Great video! I had a question about the IIA assumption and the use of probit models. You stated that standard errors are pretty large even with big samples. Is this the case even when you are working with over 100,000 cases? Thanks for all the work you put into this video.

[@bhavyasharma1256]

Hey, Doug! Thanks a lot for a fantastic video! Helped me understand the concept well. I am curious if we can find demand using Multinomial Logistic Regression? For example, demand of Strawberry flavored ice-cream in a particular area given that users have 3 options to chose from (as explained in your video)

[@ishananand4556]

Hi. I have a question. Can one use logit if the dependent variables are not mutually exclusive? Suppose we are looking at households who take debt from formal sources and informal sources, so that is 0 and 1 there, but there are households who take loans from both sources together. What kind of model can we use for this to be the dependent variable? Thanks

[@dougmckee673]

A binary logit would work if you were willing to have 1 signify a loan from either/both sources and a 0 for no loans at all. If you want to distinguish between the two types of loans, you could use a multinomial logit. Here you would have 4 possible outcomes: no loans, formal loan only, informal loan only, or both loans. In a multinomial logit, the outcomes have to be exclusive too.

[@ishananand4556]

Thanks. So if I drop the no loan category and have a population only of the indebted households, then I can have three categories- formal sources, informal sources and both. However, as your video suggests, the underlying assumption of IRA would hold. If I understood correctly, the assumption will suggest that the preference regarding the source of loan does not change with the presence of the other alternative. I am not sure whether this assumption will hold good in this case.

[@nortongartino4602]

@@ishananand4556 Hi, Anand. I think you'll find the paper titled "On the Relevance of Irrelevant Alternatives" by Benson, Kumar and Tomkins (2016) pretty helpful. It explains the IIA assumption and how nested logistic regression, which literally does not have any assumption, can be an alternative to multinomial if the data violates IIA. Cheers.

[@teshomekefale645]

hi i got good idea on the lectue introduction to multinomial Logit, but I want to know how mixed logit model function to the valuation of eco system would yuo help me

[@lanikim8765]

42:00 IIA

[@elenagarcia6334]

Thank you, you were very helpful!

[@danielsilverio6439]

Many thanks!

[@hammadabidkazi]

Thank you.

[@31carlosrivera]

Great video, but i have a question. When you are talking about predicting probabilities in the minute 33-37 shouldn´t the probabilities of ses sum 1. Thanks

[@dougmckee673]

+carlos ivan rivera Sorry for the delay--Not sure why I didn't see this when you posted it, but I'm reviewing these comments now because I'm teaching the multinomial logit in class tomorrow! Short answer: No Longer answer: These are the probabilities of each group choosing chocolate and their is no reason they should sum to 1. e.g., imagine they all have the same preferences and hate chocolate: The probability would be zero for each group and it would sum to zero. For any particular group, you could predict the probabilities of choosing chocolate, strawberry, and vanilla. THESE would sum to 1.

[@ricardpunsola]

10/10

[@dpnast8301]

Pot... is your.... enemy.... but thanx anyway :)

[@idontreallylikeyoutube]

Hi, if you invested in a better microphone, your videos would be better.

[@rafeyimane1916]

Can I have your email please?

[@xyzant5069]

could it be more attracting? almost fall asleep