Home / Topics / Research Design / Statistical Analysis / Selecting Statistical Analyses / Implementing the Institute of Medicine's Racial Disparity Definition

Implementing the Institute of Medicine's Racial Disparity Definition

Posted on December 5, 2011

Benjamin Cook examines the steps required to implement the Institute of Medicine's racial disparity definition.


So the implementation of this -- thinking really operationalizing this definition -- is you run a regression model. You put all the variables just like we did before, just like we were discussing before. You have all the variables on the right-hand side of the model, all the independent variables that you think are important to your dependent for any mental health care or something.

You put them all in the model, and you model what you think any mental health care, the kind of... You do the best job that you can in trying to understand any mental health care and all the things that go into that.

And, normally, we would look at the race coefficient and stop there. So that's sort of what's been commonly done. And so what we suggest is kind of doing something else to this. What we suggest is you get the coefficients from that model, all the betas in that regression model.

You get that from estimating the model. And then you make the African Americans and Latinos and the minority groups look like whites on health status variables. And you can do that through a... by ranking them and then replacing the lowest-ranked African American with the lowest ranked whites' health status values and kind of doing that up through the distribution until the two distributions of health status look identical between the whites and the minority group.

So that's sort of the second step. The first step is the model. The second step is some way of making the health status distributions look the same across the groups. And then the third step is to create predictions.

So you can use the model. If you just run the model through, you can generate predictions for every individual, using the coefficients from the original model, their new health status values for the minorities. We've changed their health status values. So run the coefficients from the model, their new sort of adjusted health status values and their own SES values, and you get a prediction. You get a predicted probability of whether they'll use mental health care or not.

So in the end, you'll have, for every individual, their predicted probability, their risk or their percentage probability of seeing a mental health care provider. And then you can aggregate that out by groups and then take the mean. And so it's that technology, that second and third step, the rank-and-replace method that makes the groups equivalent on health status and then that third, the prediction, using the coefficient from the original model and then the adjusted values. That technology is somewhat easily implementable, but certainly there's no... it's not a line of code in SAS or Stata or something like that.

Viewing Preferences




Excerpted from an interview with the researcher conducted at the 2011 NHSN Conference held in Miami, FL.


Please note that the feedback is viewed only by 4researchers staff and is not intended for communication with individual contributors.


Use the form below to submit feedback about this article. If you would like a response, please be sure to include your e-mail address.

More About "Selecting Statistical Analyses"


Analysis in Randomized Trials

Analyzing Count Data

Show All...


More From Benjamin LĂȘ Cook (bio)


Combining Cross-Sectional and Longitudinal Data


Equalizing Variables Across Groups

Show All...