# Random Effects Regression vs. Fixed Effects Regression

#### Veronica

##### New Member
I'm having trouble distinguishing when to use random effects versus fixed effects regression. Any thoughts?

#### bugman

##### Super Moderator
Random effects are generally used if the levels of the independent variable are thought to be a small subset of all possible values. For example, if were looking at growth rates of a plant on five randomly selected types of soil of all the soil types. In this case you may be just wanting to know if growth rates differed depending on soil.

With fixed effects, you are fixing your independant variable: so that you are specifically interested in certain soils. In this case you chose, or set the types of soil. Now your question might become more specific relating to specific soils.

This is pretty simplistic but I hope it helps.

Was are the variables you are looking at?

#### Veronica

##### New Member
Simple examples are always welcomed. :tup:

I'm studying multiple hospitals, but have patient data on hip surgery. My independent variable is focused care (continuous variable) and the dependent variable is day till patient can get out of bed post-surgery.

If I understand you correctly, then I want to know how different types of focused care that the different hospitals deliver affects the patient's mobility after hip surgery.

I don't think I understand what you mean by this: "the levels of the independent variable are thought to be a small subset of all possible values"

#### bugman

##### Super Moderator
RE: the levels of the independent variable are thought to be a small subset of all possible values

So, if you want to know say, if survival (or hip care) was dependant on the hospital they were treated at; you might not have the time, money or resources to sample all the hospitals in your city, state, country or whatever so you draw a subset of these from all the possible hospitals (i.e. the popualtion).
This is a case of treating it (hospital) as a random effect.

On the other hand, if you specifcally wanted to know if Hospital A was different from hospital B and so on - this would be an example of a fixed effect (i.e. you only care about these and none of the others in the popualtion).
Hope this clarifies it a little.

CAn you clarify focused care? How is it measured?

#### Veronica

##### New Member
Focused care is a composite measure that we created on a scale of 1 to 10. The indicator represents the extent to which the hospital dedicates staff time, facility space, , technology, and protocols for hip surgery. The more resources focused on hip surgery, the more specialized the hospital. ...And you would hypothesize, that this would also mean better mobility outcomes for the patient because the care is super specialized.

:yup: I understand your explanation perfectly, but now I'm uncertain if I should be using random effects or fixed effects. I have 21 hospitals, supposedly a subset of the universe of hospitals, but I do want to tease out the effect of the different degrees of focus in these hospitals on the outcome. I don't necessarily want to know the effect of focus for these 21 hospitals, but for the universe of hospitals with degrees 1 through 10.

Now that I've said that, it is random effects that I should do, isn't it?

#### bugman

##### Super Moderator
Treat Hospital as a random effect if its in your model and focused care as a fixed effect since it has boundaries and you are specifically interested in the various levels.

#### Veronica

##### New Member
Ah, I see. Thanks for your help Bugman!