发帖
雷达卡
We are using HLM Version 7 to analyze the NAEP NIES restricted data sets for 2009 and were running into some questions. We are analyzing a series of models starting with an unconditional model (Model 1) and then adding variables at level one (Model 2), level 2 (Model 3), and level 3 (Model 4). Each model incorporates 5 plausible values as an outcome and we are using sampling weights. We have no missing values in the data sets we are using, and have also tested assumptions and found no problems.
- Looking at the percent of the variance accounted for across models, the total goes down as we add variables as expected. However, the percent variance accounted for within each level fluctuates. The actual numbers look similar to the ones in the table below with the total variance going down as we add predictors but at some levels the variance increases (e.g., level three variance between Model 2 and Model 3 and level one between Model 3 and Model 4). Does anyone have any thoughts why this might be happening?
- Model 1 Unconditional
- Model 2 Level one covariates added
- Model 3 Level two covariates added
- Model 4 Level three covariates added
Model | 1 | 2 | 3 | 4 |
Level one variance | 620 | 520 | 520 | 540 |
Level two variance | 210 | 300 | 150 | 150 |
Level tree variance | 130 | 130 | 260 | 120 |
Total | 960 | 950 | 930 | 810 |
- We are also noticing that the degrees of freedom for some level-1 fixed effects are much smaller than we expected while some are fairly large. Is this because of the method that the HLM software uses to calculate degrees of freedom when plausible values are used?
Any assistance or ideas are greatly appreciated.
京公网安备 11010802022788号
