Logistic Models in R

Loading...

From the course by University of Maryland, College Park

Combining and Analyzing Complex Data

26 ratings

University of Maryland, College Park

Combining and Analyzing Complex Data

26 ratings

Course 6 of 7 in the Specialization Survey Data Collection and Analytics

In this course you will learn how to use survey weights to estimate descriptive statistics, like means and totals, and more complicated quantities like model parameters for linear and logistic regressions. Software capabilities will be covered with R® receiving particular emphasis. The course will also cover the basics of record linkage and statistical matching—both of which are becoming more important as ways of combining data from different sources. Combining of datasets raises ethical issues which the course reviews. Informed consent may have to be obtained from persons to allow their data to be linked. You will learn about differences in the legal requirements in different countries.

From the lesson

Models

Module 2 covers how to estimate linear and logistic model parameters using survey data. After completing this module, you will understand how the methods used differ from the ones for non-survey data. We also cover the features of survey data sets that need to be accounted for when estimating standard errors of estimated model parameters.

Logistic Models in R4:53

Odds Ratios8:50

Meet the Instructors

Richard Valliant, Ph.D.
Research Professor
Joint Program in Survey Methodology

In this video we will illustrate how to fit a logistic model in R.

So I'm going to use the same data set that we've seen before,

this academic performance data set that comes bundled with the R survey package.

We will look at a logistic model to predict whether a school

met the target for school-wide growth in the API score.

Now in other words, did they improve their scores compared to the previous year?

And then we will see how to test whether subset of coefficients is zero,

the same way we did in the linear model.

And we'll also look at odds ratios for

a categorical predictor to see how that goes.

So there's a number of variables in the API dataset and

here are the ones that we're going to deal with in this illustration.

School, sch.wide, is just an indicator, did the school meet the growth target?

No or yes.

ell, meals, mobility we’ve met before in previous videos.

Enrollment is the number of students enrolled in the school.

hsg is the percentage of parents who are high-school grads,

the more educated the parents, the better the students may be doing.

Then there's col.grad, that's percentage of parents with college degree.

And then also there's an indicator, yes or no, for

whether the school is a year-round school or not.

Do they take off in the summer or do they go 12 months during the year?

So here's the R code.

Now some of this we've seen before just to get going.

We require the package, tell it what data we're going to use,

and then create a design object.

Those statements we've seen before.

We, once again, use the svyglm function to fit the model.

This is the same function we used for the linear model.

So we set up a formula here for,

School-wide as predicted based on all the variables in that last table.

And the thing that's different here is we use the family parameter, so

we specify that as quasibinomial link = logit.

That leads to logistic regression.

If you put probit in here then you get probit, and

cloglog would give you complimentary log-log.

So here's the output.

I do a summary on m2, which is where I saved the model result.

And what have we got?

We've got one variable, enrollment, highly significant.

We've got two here that are significant at the 10%

level, mobility and percent college graduates.

So the other things, though, don't appear to be that important.

Year-round school, close to 10% level significance but doesn't quite make it.

And then the other things are test that is being fairly different from 0.

Or not testing out is being fairly different from 0.

Now we can test a subset of coefficients the same way we did in the linear model

using the regTermTest function.

So I specify in my model here, after tilde

I specify the set of parameters I want to test.

So I'm doing English language learners percentage with subsidized meals and

percent of parents with high school grads.

And I want to do a Wald test on that.

So It echoes back what I ask it to do and

gives me the F statistic as 0.2395 on 2 and

190 degrees of freedom.

p value is 0.8686, so that's highly non-significant.

So it looks like the evidence is pointing to all three

of those parameters simultaneously being 0 in this model.

So in the next video we'll follow up with how

to estimate odds ratios in a logistic model.

Explore our Catalog

Join for free and get personalized recommendations, updates and offers.

Coursera provides universal access to the world’s best education, partnering with top universities and organizations to offer courses online.

© 2018 Coursera Inc. All rights reserved.

Download on the App Store

Get it on Google Play