understanding the prevalance of covid

On Twitter, Gabriel Rossman is interested in the issue of selection bias in a recent study of covid. What happens if a study is contaminated by selection bias?

Excellent point. But this is a problem with all surveys – not only do you have to randomly sample people, they have to randomly respond. Fortunately, there are ways to understand how survey participation biases estimation.

Strategy 1: If you have data on responders and non-responders (e.g., SES data on neighborhood, some basic demographic data), you can create a two stage model. This is the intuition behind the Heckman model.

Strategy 2: If you have some intuition on why people do not respond, you can simulate non-response and create bounds on your errors.

Strategy 3: Go full Bayesian. Admit that bias exists in the model but build that into an update on what you think the true parameters are.



50+ chapters of grad skool advice goodness: Grad Skool Rulz ($4.44 – cheap!!!!)
Intro to sociology for just $1 per chapter – INSANE BARGAIN!!!!!
A theory book you can understand!!! Theory for the Working Sociologist (discount code: ROJAS – 30% off!!)
The rise of Black Studies:  From Black Power to Black Studies 
Did Obama tank the antiwar movement? Party in the Street
Read Contexts Magazine– It’s Awesome!!!!

Written by fabiorojas

April 22, 2020 at 6:14 pm

Posted in uncategorized

%d bloggers like this: