[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Re: Anova of clustered data

From   David Airey <>
Subject   Re: st: Re: Anova of clustered data
Date   Tue, 19 Feb 2008 08:25:41 -0600


Are the 25 wires and 5 brackets replicates or different types or wires and brackets? I'm guessing your goal is to estimate the average wire load, but that you don't want other factors or covariates to bias that estimate (CI) of average load. In this case, you could take advantage of xtmixed to get a variance component model to account for random variation associated with bracket and wire, and look to the model constant for the estimate of average load that you want. You could even add a time covariate if you measured that.

I'm not sure the t-test is doing what you want to do, which seems to be to estimate the wire load. With t-tests you are asking if there are differences between whatever two groups you considered. To me it sounds like you want to estimate wire load accounting for: variance associated with different wires, variance associated with different brackets, and measured time of the pull. If you did not measure the latter, it becomes part of the random error associated with wire.

I recommend "Multilevel and Longitudinal Modeling Using Stata", 2nd Edition by Rabe-Hesketh and Skrondal, from the Stata book store. The second edition is brilliant!


On Feb 19, 2008, at 3:46 AM, Paul Fenner wrote:

I have an experiment in which a wire is drawn through a bracket and the load is measured as a function of time over a constant time period. This is repeated for 25 different wires in each bracket and there are 5 brackets, and I want to see if the brackets are different. Ideally the load should be independent of time but it fluctuates due to the nature of the experiment - friction etc.
I am concerned that for each wire the loads may not be strictly independent and that the data should be treated as clustered for each wire.
I could not firure out how to allow for this in an anova desighn so I used a t-test with the vce(cluster) option for all possible combinations and then multproc and smileplot to analyses the multiple comparisons.

Is there a better (more efficient) way of carrying out this analysis?
Free games, great prizes - get gaming at Gamesbox.

* For searches and help try:
*   For searches and help try:

© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index