I'm deeply stuck in an issue in my research paper's methodology. I'm getting the following disparate results for my marginal effects.

  1. In my Ordered Logit model, I have Y = Trust in banks (4-Max trust, 3-High Trust, 2- Low Trust, 1-Least trust). Here are my marginal effects using ocME (the SS is probably attached in the post with the effect1, effect 2, etc headings, and it's the model with the name two_dummies_model.)

Ordered Logit Model: Marginal Effects

I have attached another screenshot of the summary of this model. As you can see, the signs of the values (log-odds) are opposite e.g. Religion1's ME was negative but the value in the summary is positive. Why is this so? What am I doing wrong.

Summary of Ordered Logit Model

Secondly, to diagnose this issue, I have then transformed the data so that the Y variable is now binary (grouped levels 3 and 4 into 1 and levels 1 and 2 into 0. Now 0 indicates Low/No trust and 1 indicates High trust). With a binary Y variable, I have now used logistic regression (glm) instead of ordered logit. Below is the summary of the marginal effects (model name is twoD_model_log.) The 0 values are those variables which I have not included in my regression equation.

Marginal Effects of Logistic regression

As you can see if you compare the MEs of these two models,, the marginal effects of most variables are of opposite signs e.g. female marginal effect is -0.03 in logistic model and 0.012 (effect1) in Ordered logit model. Any idea why this is so? And which model should I use?

I am very confused in this. Spent hours trying to figure it out but no luck. Even CGPT doesn't help :(



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.