Abstract
Experiments with repeated measurements are common in pharmaceutical trials, agricultural research, and other biological disciplines. Many aspects of the analysis of such experiments remain controversial. With increasingly sophisticated software becoming available, e.g. PROC MIXED, data analysts have more options from which to choose, and hence more questions about the value and impact of these options. These dilemmas include the following. MIXED offers a number of different correlated error models and several criteria for choosing among competing models. How do the model selection criteria behave? How is inference affected if the correlated error model is misspecified? Some texts use random between subject error effects in the model in addition to correlated errors. Others use only the correlated error structure. How does this affect inference? MIXED has several ways to determine degrees of freedom, including a new option to use Kenward and Roger's procedure. The Kenward-Roger procedure also corrects test statistics and standard errors for bias. How do the various degree-of-freedom options compare? When is the bias serious enough to worry about and how well does the Kenward-Roger option work? Some models are prone to convergence problems. When are these most likely to occur and how should they be addressed? We present the results of several simulation studies conducted to help understand the impact of various decisions on the small sample behavior of typical situations that arise in animal health and agricultural settings.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.
Recommended Citation
Guerin, LeAnna and Stroup, Walter W.
(2000).
"A SIMULATION STUDY TO EVALUATE PROC MIXED ANALYSIS OF REPEATED MEASURES DATA,"
Conference on Applied Statistics in Agriculture.
https://doi.org/10.4148/2475-7772.1249
A SIMULATION STUDY TO EVALUATE PROC MIXED ANALYSIS OF REPEATED MEASURES DATA
Experiments with repeated measurements are common in pharmaceutical trials, agricultural research, and other biological disciplines. Many aspects of the analysis of such experiments remain controversial. With increasingly sophisticated software becoming available, e.g. PROC MIXED, data analysts have more options from which to choose, and hence more questions about the value and impact of these options. These dilemmas include the following. MIXED offers a number of different correlated error models and several criteria for choosing among competing models. How do the model selection criteria behave? How is inference affected if the correlated error model is misspecified? Some texts use random between subject error effects in the model in addition to correlated errors. Others use only the correlated error structure. How does this affect inference? MIXED has several ways to determine degrees of freedom, including a new option to use Kenward and Roger's procedure. The Kenward-Roger procedure also corrects test statistics and standard errors for bias. How do the various degree-of-freedom options compare? When is the bias serious enough to worry about and how well does the Kenward-Roger option work? Some models are prone to convergence problems. When are these most likely to occur and how should they be addressed? We present the results of several simulation studies conducted to help understand the impact of various decisions on the small sample behavior of typical situations that arise in animal health and agricultural settings.
