A practical and example filled tour of simple and multiple regression techniques (linear, logistic, and Cox PH) for estimation, adjustment and prediction.

Loading...

來自 Johns Hopkins University 的課程

Statistical Reasoning for Public Health 2: Regression Methods

81 個評分

A practical and example filled tour of simple and multiple regression techniques (linear, logistic, and Cox PH) for estimation, adjustment and prediction.

從本節課中

Module 2A: Confounding and Effect Modification (Interaction)

This module, along with module 2B introduces two key concepts in statistics/epidemiology, confounding and effect modification. A relation between an outcome and exposure of interested can be confounded if a another variable (or variables) is associated with both the outcome and the exposure. In such cases the crude outcome/exposure associate may over or under-estimate the association of interest. Confounding is an ever-present threat in non-randomized studies, but results of interest can be adjusted for potential confounders.

- John McGready, PhD, MSAssociate Scientist, Biostatistics

Bloomberg School of Public Health

Greetings!

In this next lecture set,

we're going to more formally revisit the concept of confounding.

This is an idea we talked about in Statistical Reasoning 1, and

now we're going to formally define it in terms of the mechanisms that cause

confounding to occur, and we're going to start talking about how to adjust for

confounding when associating two variables of interest.

So, the idea of confounding occurs when we're trying to look at some kind of

two-variable association.

We'll say, outcome exposure, and there's other factors that are related

to the outcome exposure that may either magnify the actual association or hide it.

And if we don't account for

these other factors, we may miss the story about our y-x relationship of interest.

So, we're going to talk about in this section here,

what conditions are necessary for confounding,

what can be done in study design to minimize the threat of confounding, and

how to deal with confounding if it is a possibility in your data?

And so, we're going to talk about the idea of confounder-adjusted estimates

of association.

Associations between an outcome and

a predictor that were moved behind the scenes.

Associations with other variables that may be magnifying or

hiding the actual association.

And this is going to be your first talk of the idea of adjusted estimates, so

soon when we get into multiple regression, we'll see that we

can actually estimate outcome exposure relationships adjusted for

potential compounders, very easily in the framework of the multiple regression.

Sometimes, in epidemiological terms, confounders are called lurking variables.

Variables that are behind the scenes, and

related to the things we're trying to associate.

So, it's very fitting today is Halloween 2013, and I'm recording this video, and so

maybe a good costume idea, if you're, still need one for

next year, is to be a confounder, a lurking variable.

It can be a scary thing, at least in terms of data analysis, right?

So anyway, enough about my costume ideas.

Onward and upward and we'll talk formally about this idea of confounding.