Trenton Mize

@richard-williams.bsky.social is an angel and made a minor update to golgit2 so that it works smoothly with meinequality and totalme.

So, just update to the newest gologit2:

adoupdate gologit2, update

And you should be good to go.

Love it. Better than an actual donut chart too: www.jaspersoft.com/articles/wha...

But it was a really good idea!

Do check out this super cool paper on how to summarize effects for nominal dependent and independent variables, published in the one and only Sociological Science! 👇 #sociology

Indeed! A secret methodologists don't tell everyone is that we usually create new methods out of selfishness for solving annoying problems we face in our own work 😅.

For Stata users, we wrote two new commands that implement our methods: MEinequality and totalME www.trentonmize.com/software. For R users, we provide template code which implements our methods: www.trentonmize.com/research/miz...

ME inequalities and total MEs have myriad applications. They can summarize effects, they enable comparisons of effects sizes across variables, they simplify tests of interactions, they can be used in tests of mediation, they can be used to compare effects across groups, and more.

Effects of age and marital status on self-rated health. Although self-rated health has four categories, we can use total marginal effects to summarize the effect of each predictor on the outcome.

With nominal/ordinal outcomes, we are often interested in the overall effect on the outcome rather than effects on each category. For example, how age or marital status affect self-rated health as a holistic construct. We develop a total marginal effect (ME) measure as a summary measure.

Hypothetical example of two nominal independent variables. Nominal in right panel has much larger effect as the predictions are more spread out, showing it patterns more unequal outcomes

Does the nominal independent variable (IV) in the left or right panel have a larger effect? Intuitively, it is the one on the right because it patterns more unequal outcomes. We develop a marginal effect (ME) inequality statistic which summarize the holistic effect of a nominal or ordinal IV.

Are you tired of analyzing your nominal and ordinal variables like its the 1950s? Then read @sociologicalsci.bsky.social today and see if ME inequality and total ME are right for you. We develop new methods for summarizing effects for nominal/ordinal independent and dependent variables.

You even made the acknowledgements!

Graph showing predicted mental health across the range of obligatory roles held. For younger and middle-aged adults, more obligatory roles predicts better mental health. In contrast, there is no effect of obligatory role counts on mental health for the oldest adults.

For more intensive obligatory roles, role-accumulation is beneficial only for the young and middle-aged. But there is a silver lining here: older adulthood is associated with obligatory role loss, and the null findings for them suggest resilience and strong mental health in the face of role loss.

Graph showing predicted mental health across the range of voluntary roles held. For all ages, more voluntary roles predicts better mental health. But the effect is strongest for older adults who see strong mental health benefits to holding more voluntary roles.

As sociological theories of identity and role-accumulation predict, voluntary roles are beneficial for mental health at all ages. In contrast to some theories of aging, they are even more beneficial for the oldest adults who tend to hold fewer social roles overall.

Excited about a new article with Reilly Kincaid in ASR @asanews.bsky.social on role-accumulation & mental health across the life course. We identify a disagreement across prominent theories for how role-accumulation should affect mental health, especially in older adulthood. doi.org/10.1177/0003...

I made one for stats papers

Love a good hot take. In my experience the coefficients across the two almost never differ meaningfully. But those are effects on the rate/count. You often do see quite different predictions for the observed count (e.g. Pr(y=0)) and Poisson is usually very wrong and NB much closer in these cases.

Thanks for making this. Can you please add me?

This is so useful! Please add me 😊.

I made this to teach R^2. Pretty proud of myself.

Many of these methods use under the hood…a logit 😋.