Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Stat Procs
- /
- Mixed modeling

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

02-23-2011 08:47 AM

1. Generating Data from Mixed Models and Estimating parameters using PROC MIXED

Consider a situation of Mixed Models where one gets n observations from k groups. Assume that response variable Y follows the mixed model

Yij = μ + βi Χij + εij i = 1,...k; j = 1,...,n,

where μ is a fixed common effect, βi is a random effect (common for a given group and different for different groups) and εij are error terms. Assume that the all random effects are normally distributed as

βi ~ N(0, σb2 ) and that εij ~ N(0, σe2 )

Assuming that

(a). Χij = 1 for all i and j as in a simple random effects model, k=5 and n=10, using SAS rand('normal') function generate a set of Yij when μ =10, σb = .5 and σe = 1.

(b). Using the above data estimate parameters μ and βi as well as σb = .5 and σe parameters by PROC MIXED and compare estimates with actual values (note: they can be very different due to small sample size, but we will study average performance below).

(c) Repeat above two steps when the covariate vector for each group is a vector from 1 to 10; i.e. Χi. = c(1,2,....n).

2. Simulation: Testing Accuracy of SAS Estimates

Convert the data generation in Exercise 1 above into a macro with σb and σe as parameters. Under the above two scenarios on covariate X, call your macro 1,000 times and report

(a) average of βi estimates when σb = .5 and σe = 1,

(b) average of σe estimates when σb = .1 and σe = 1,

(c) average of σb estimates when σb = .1 and σe = 1.

Can anyone help me on this.

Thanks,

Siddhartha

Consider a situation of Mixed Models where one gets n observations from k groups. Assume that response variable Y follows the mixed model

Yij = μ + βi Χij + εij i = 1,...k; j = 1,...,n,

where μ is a fixed common effect, βi is a random effect (common for a given group and different for different groups) and εij are error terms. Assume that the all random effects are normally distributed as

βi ~ N(0, σb2 ) and that εij ~ N(0, σe2 )

Assuming that

(a). Χij = 1 for all i and j as in a simple random effects model, k=5 and n=10, using SAS rand('normal') function generate a set of Yij when μ =10, σb = .5 and σe = 1.

(b). Using the above data estimate parameters μ and βi as well as σb = .5 and σe parameters by PROC MIXED and compare estimates with actual values (note: they can be very different due to small sample size, but we will study average performance below).

(c) Repeat above two steps when the covariate vector for each group is a vector from 1 to 10; i.e. Χi. = c(1,2,....n).

2. Simulation: Testing Accuracy of SAS Estimates

Convert the data generation in Exercise 1 above into a macro with σb and σe as parameters. Under the above two scenarios on covariate X, call your macro 1,000 times and report

(a) average of βi estimates when σb = .5 and σe = 1,

(b) average of σe estimates when σb = .1 and σe = 1,

(c) average of σb estimates when σb = .1 and σe = 1.

Can anyone help me on this.

Thanks,

Siddhartha