ohh
Full Member
Posts: 224
|
N=crap
Dec 6, 2011 14:19:29 GMT -5
Post by ohh on Dec 6, 2011 14:19:29 GMT -5
So I am collecting survey data on patients getting a specific procedure in a family planning clinic that got it's funding cut thanks to the Republican legislators in the state.
Right now I have 60 surveys. Due to the cuts, there will only be about 1 patient a week coming in. Whereas it won't involve that much time (I work in the hospital) for me, I am wondering if it's worth it to keep getting surveys.
I mean, is there really a difference between 60 and 70 survey respondents?
|
|
|
N=crap
Dec 6, 2011 15:23:24 GMT -5
Post by drbearjew on Dec 6, 2011 15:23:24 GMT -5
I've seen N of <30 justified in some survey research. You're fine.
|
|
|
N=crap
Dec 10, 2011 22:45:49 GMT -5
Post by unclekarl on Dec 10, 2011 22:45:49 GMT -5
It depends on what you are looking for. You can always estimate statistical power to see the amount of standard error for any quantitative variables you are estimating.
|
|
|
N=crap
May 6, 2015 10:10:56 GMT -5
Post by depends on May 6, 2015 10:10:56 GMT -5
I think it depends what you are doing with the data. If you are doing stats on it, it might be worth going for more. Above 50 is ok but once you start to slice and dice it is better to have more. Difference between 60 and 70 can mean the difference between significance or not
|
|
bigger is better, but...
Guest
|
N=crap
May 6, 2015 11:07:52 GMT -5
Post by bigger is better, but... on May 6, 2015 11:07:52 GMT -5
The first rule of sample sizes is to always do a power analysis (unclekarl is right here). What do you expect effect sizes to be? Is your sample big enough to detect that?
However, part of this is that it also depends on how many parameters you have in the analysis. You're going to eat up a lot of degrees of freedom (relative to the sample size) the more parameters you have in your model. The difference between 60 and 70 isn't as dramatic as between 40 and 50, but it's not like the difference between 200 and 210 (which isn't that much of a difference). In a simple OLS, the "magic number" is usually about 120 degrees of freedom, since that's where the t-distribution starts looking like a z-distribution.
If it's not too much trouble, I would urge getting to 70 if possible. But power analyses will help you a lot here.
|
|