marginal effects in logit model

[Isaac Kimunio]

Dear members,

Currently I am struggling with doing logit regression using panel data. Am using a panel of banks for 10 years.
I am regressing FSIIndex (1 if index is higher for all banks including foreign banks and 0 if index is higher for banks excluding foreign banks) Loans Loansgrowth non-performingloans GDP Interestrate.
FSI is financial stability index.
My questions are:
1. Do I use the binary outcomes logit or the ordinal outcomes logit?

• If I carry out a fixed effect logit I get the following

xtlogit FSIIndex LnLOANS Loansgrowthall Loansgrowthforeign LD LnNPLs GDP r, fe
note: multiple positive outcomes within groups encountered.

• . mfx

default predict() is unsuitable for marginal-effect calculation

• When I do a random effect I get the following:

xtlogit FSIIndex LnLOANS Loansgrowthall Loansgrowthforeign LD LnNPLs GDP r, noconstant revce(robust)

Calculating robust standard errors:
conformability error

When I do an ordinal outcome logit

xtologit FSIIndex LnLOANS Loansgrowthall LnNPLs LD GDP r, vce(robust)

. mfx

Marginal effects after are exactly same as coefficients

2. How do I get the marginal effects after that?

3. Which logistic regression is more correct and what is the difference of each regarding the marginal effect postestimation

[Phil Bromiley]

You'll increase your chances of a helpful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

Your description seems like a binary dv, so I don't see any reason for ordinal logit. If you have more than two ordered outcomes, ordinal logit looks appropriate.

I tend to margins rather than user-written mfx. Predictions in conditional logit may need to assume only one positive outcome per panel which is not the case in your data.

If this is what you ran: xtlogit FSIIndex LnLOANS Loansgrowthall Loansgrowthforeign LD LnNPLs GDP r, noconstant revce(robust)

I'm surprised it ran at all - no space between re and vce. What if you do cluster?