|
|
 |
| Minimum level-1 sample size? |
 
|
| Message/Author |
|
|
| Jane W. posted on Wednesday, March 12, 2014 - 3:15 pm
|
|
|
I have data on 200+ employees who provided twice-daily ratings over a 7-day period. Which I see as:
- Level 3 (employee): n=200+
- Level 2 (day): n=7
- Level 1 (event/within-day): n=2
The variance is mostly distributed across levels 1 and 3, with anywhere from 0-8% of variance occurring at level
2.
Alternatively, I could run a 2-level model and collapse across day, such that:
- Level 2 (employee): n=200+
- Level 1 (event): n=14
Which would help if level-1 fixed effects were of primary interest.
My questions:
1) In general, is n=2 too small for level 1?
2) Would you run as a 2- or 3-level model?
Thanks for any input! |
|
|
|
|
You have too few level 2 units for three-level analysis. You should use alternative 2. You need at least 20 and some would say 30-50 clusters.
N=2 is not too small for level 1. |
|
|
| Back to top |
|
|
