Return to Question

• Is this the right approach to meet my aim of partitioning variance among the two levels of REM status (White/REM) and the two levels of analysis (patient/therapist)? In the end, I am hoping to be able to calculate variance explained across all 3 levels (patient, provider residual) by 3 samples (total sample, white, rem) for a total of 9 ICCs.

• Is this the right approach to meet my aim of partitioning variance within the two levels of REM status (White/REM) and across the two levels of analysis (patient & provider)? In the end, I am hoping to be able to calculate variance explained across all 3 levels (patient, provider residual) by 3 samples (total sample, white, rem) for a total of 9 ICCs.

I remain with a few questions:

• How do I interpret the Groups under Random effects? e.g., rem:id versus id ... It looks like there are two intercepts for my level-1 variables, but I do not know if rem:id is across rem status and id is across all patients. If that is the case I do not believe this model is right for my question. I am looking for something like white:id and rem:id and maybe that is not possible with this model.
• The number of observations looks mostly right, however, for rem:provider the number is double the number of providers in my sample. And rem:id is equal to total number of patients as well as id. In my sample White is about 1000 and REM is about 500. Is this indicative of an error in my model, or is it not assessing these two groups separately?
• My residual variance across both models looks to be forced into one variance estimate. Is there a way to partition this variance across level-1 and level-2? This will be critical in calculating ICCs.
• Slight aside: but do I need to include rem as a fixed effect as well? How does including versus not including impact variance estimation?

In the end, I am hoping to be able to calculate variance explained across all 3 levels (patient, provider residual) by 3 samples (total sample, white, rem) for a total of 9 ICCs.

Central Question:

• Is this the right approach to meet my aim of partitioning variance among the two levels of REM status (White/REM) and the two levels of analysis (patient/therapist)? In the end, I am hoping to be able to calculate variance explained across all 3 levels (patient, provider residual) by 3 samples (total sample, white, rem) for a total of 9 ICCs.

I struggle to interpret the Groups under Random effects... e.g., rem:id versus id ... It looks like there are two intercepts for my level-1 variables, but I do not know if rem:id is across rem status and id is across all patients. I am looking for something like white:id and rem:id to estimate my ICCs.

Secondary Model Results Questions:

• The number of observations looks mostly right, however, for rem:provider the number is double the number of providers in my sample. And rem:id is equal to total number of patients as well as id. In my sample White is about 1000 and REM is about 500. Is this indicative of an error in my model, or is it not assessing these two groups separately?
• My residual variance across both models looks to be forced into one variance estimate. Is there a way to partition this variance across level-1 and level-2? This will be critical in calculating ICCs.
• Do I need to include rem as a fixed effect as well? How does including versus not including impact variance estimation?

Model 2 output is as follows:  REML criterion at convergence: 13694.2 Scaled residuals: Min 1Q Median 3Q Max -10.2143 -0.3806 0.0751 0.4585 5.6865 Random effects: Groups Name Variance Std.Dev. rem:id (Intercept) 0.21617 0.46494 id (Intercept) 0.36515 0.60428 rem:provider (Intercept) 0.05808 0.24100 provider (Intercept) 0.00515 0.07176 Residual 0.17966 0.42386 Number of obs: 8606, groups: rem:id, 1504; id, 1504; rem:provider, 78; provider, 39 Fixed effects: Estimate Std. Error df t value Pr(>|t|) (Intercept) 5.9928132 0.0483799 66.1700000 123.870 <2e-16 *** remrem -0.0005091 0.0740942 52.5400000 -0.007 0.995 --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Correlation of Fixed Effects: (Intr) remrem -0.615 

Model 2 output is as follows:

 REML criterion at convergence: 13694.2 Scaled residuals: Min 1Q Median 3Q Max -10.2143 -0.3806 0.0751 0.4585 5.6865 Random effects: Groups Name Variance Std.Dev. rem:id (Intercept) 0.21617 0.46494 id (Intercept) 0.36515 0.60428 rem:provider (Intercept) 0.05808 0.24100 provider (Intercept) 0.00515 0.07176 Residual 0.17966 0.42386 Number of obs: 8606, groups: rem:id, 1504; id, 1504; rem:provider, 78; provider, 39 Fixed effects: Estimate Std. Error df t value Pr(>|t|) (Intercept) 5.9928132 0.0483799 66.1700000 123.870 <2e-16 *** remrem -0.0005091 0.0740942 52.5400000 -0.007 0.995 --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Correlation of Fixed Effects: (Intr) remrem -0.615