This always bothers me: don't use an exotic model if the results don't really change from the simple OLS estimator. MLE is just nonsensical compared to OLS. Sure, binary response, use logit or probit. But, ordered logit? Waste of time. Parallel regression assumption is always violated. Just run it in OLS  significance levels are almost always, in my experience, unchanged and you get the added benefit of interpretable partial slopes. OLS is striking in that it's usually robust to what we call "misspesifications." Don't reinvent the wheel, people.
Just use OLS

This always bothers me: don't use an exotic model if the results don't really change from the simple OLS estimator. MLE is just nonsensical compared to OLS. Sure, binary response, use logit or probit. But, ordered logit? Waste of time. Parallel regression assumption is always violated. Just run it in OLS  significance levels are almost always, in my experience, unchanged and you get the added benefit of interpretable partial slopes. OLS is striking in that it's usually robust to what we call "misspesifications." Don't reinvent the wheel, people.
It's like you've just finished baby Wooldridge and you're excited to show off how smart you are now.

When you don't have anything substantive to say, you cloak your work in the most complicated and uninterpretable models possible.
Praise be to the scholar whose work is only able to be understood by 5 or 6 other people who are aware of their new method. But it's OK, they show how their new estimator pretty much exactly mimics existing estimators that thousands are familiar with.

I throw the "fancier" models in the appendix.
I have had more success with reviewers by doing the opposite. I'll have the fancy model with all of the accompanying text about the reasons why its particular set of assumptions and form best fit the theory of the data generating process in the main text. Then there is an OLS version in the Appendix to say, "By the way, the results weren't contingent on this seemingly obscure model choice. Here's an OLS showing pretty much the same thing."
If I thought it wouldn't hurt my chances with reviewers, I'd just use OLS a whole lot more often. 
^I think you can usually do the opposite, and it's what I prefer. Show the results as simply as possible, with OLS. Then do the more complicated stuff in the appendix and calculate predicted probabilities to show that it's robust to model choice. No one is going to reject your paper for that. Some day when we purge the field of all the useless garbage that methodologists have brought in we'll just be able to go back to a careful use of tabulation and graphs, especially where experiments are concerned.

I would like to believe you. But I think we are walking backward into the future.
^I think you can usually do the opposite, and it's what I prefer. Show the results as simply as possible, with OLS. Then do the more complicated stuff in the appendix and calculate predicted probabilities to show that it's robust to model choice. No one is going to reject your paper for that. Some day when we purge the field of all the useless garbage that methodologists have brought in we'll just be able to go back to a careful use of tabulation and graphs, especially where experiments are concerned.

I don't know. I hope not. One of the main reasons I support the explosion of experiments in poli sci, even though so many are piddling and dumb, is that I thought it might be a way for the methodologists to put themselves out of business.
I would like to believe you. But I think we are walking backward into the future.
^I think you can usually do the opposite, and it's what I prefer. Show the results as simply as possible, with OLS. Then do the more complicated stuff in the appendix and calculate predicted probabilities to show that it's robust to model choice. No one is going to reject your paper for that. Some day when we purge the field of all the useless garbage that methodologists have brought in we'll just be able to go back to a careful use of tabulation and graphs, especially where experiments are concerned.