Small observation in multinomial logistic regression

[samaneh khaef]

Hello
I am trying to know what region is more attractive for people. so I used region ( five categorical variables) as a dependant.
people characteristics are education (high low medium), country of origin (EU, nonEU US Africa Asia), and type (Family, Student, Worker, Refugees). All are categorical
I also aim to interact education with type fn origin. But as there are only two low and medium educated students and 5 low educated EU in one specific region, the relative risk ratio for these groups in interaction is extremely high with empty intervals. So I remove them. running my MNL, results are much better.

but as e.g. low educated students are removed for students I have two levels of education while for other groups I have three educational levels. so the low educated student is zero while high educated students are also omitted! basically, how should I exclude some observations in MNL?

[Carlo Lazzaro]

Samneh:
why not grouping together categorical levels with a small number of observations instead of getting rid of them?

[samaneh khaef]

well, I want to know how education and type separately affect the region of living and then how this combination matters. so making one category I wont be able to find such effect

[samaneh khaef]

I mean for students, there are very few who have low education. so I remove them. but now I have difficulty to interpret results as they are like:
given that I have no low educated student as low educated is a reference here, then how should I read it? region D is reference

[Carlo Lazzaro]

Samaneh:
I find this outcome table impossible to comment on.
You reported interaction but n main condictional effect of the predictors included in each interaction.
Please read the FAQ on how to share what you typed and what Stata gave you back. Thanks.

[samaneh khaef]

sorry I am now allowed to share data so I put the final result here. I modify it. my main question is as students are not low educated in my data so the reference which is low education is empty and other levels omitted. what suggestions do you have

?

[Carlo Lazzaro]

Samaneh:
I would calculate predicted probabilities for the different outcomes to make things clearer and easier to disseminate (see -predict- available from -mlogit postestimation-).

[samaneh khaef]

Thanks, Carlo
basically, I am wondering how should I run a MNL when in an interaction like here one of the categories like the student has only two educational levels while other categories as workers, refugees, and family have three educational levels?

[Carlo Lazzaro]

Samaneh:
if you omitted low educated student from your regression, this level of the categorical variable will be embedded in the residuals.
The mai conditiona effect -Student- represents high-educated student (who are probably the reference category).

[jeffrey bosah]

The effects of brexit on fdi comparing uk and EU27 countries. using difference in difference estimation technique

Today, 05:40
Hey everyone, please i have been struggling to model my DID estimation on stata for my dissertation which i have a deadline on the 5th of january. and my supervisor is not responding to his emails because his on holiday. i really need help.

so basically what i am trying to do is too understand the pre and post effect of brexit (policy change) on fdi between uk and EU27 countries.
these are what i am too achieve;
DID estimation
graphical diagnostics and test (parallel trend)
standard error considerations
default cluster-robust standard error

my data is also a longitudinal panel dataset from 2000-2019 . it is strongly balance but any time i try to model the DID estimation i always have problem with the treatment variable or group variable on stata.

please everyone willing to help personally as well my email address is [email protected], i will attached my dataset. i will really appreciate everyone help on this

[jeffrey bosah]

please do take a look at the data attached file

[Carlo Lazzaro]

Jeffrey:
as per FAQ, you should be already aware that attachments are not encouraged not to say opening spreadsheet coming from unknown sources (malware/active contents are a real risk, as we know).

[samaneh khaef]

Thanks. but if I do not omit, let's assume there is no student with low education. then how will my base category will be for this group? for others e.g. workers with high and medium education are compared with low education but as I do not have low educated students, how the comparison is done? how should I read the result

[Carlo Lazzaro]

Samaneh:
add to your -mlogit- code the -allbaselevels- option (see -help estimation options-).

[jeffrey bosah]

oh alright thanks will send a screenshot of the dataset.