Storing Weights
main topic 

You can store weights for your response using fitted or adjusted variances. Whether you use fitted or adjusted variances depends on whether you have repeat or replicate measurements. Once you have stored the weights, you can specify them in Analyze Factorial Design > Weights to perform weighted regression when analyzing the location model.  

Weighted regression is a method for handling data with observations that have different variances. If the variances are not constant, observations with:

·     Large variances should be given relatively small weight

·     Small variances should be given relatively large weight

If the variability of responses differs significantly across factor settings, consider using weighted regression if you analyze the location effects of your response.

Weights for replicates (unadjusted weights)

Store unadjusted weights using the fitted variance, if the data contain replicate measurements. Use the unadjusted weights when analyzing the location effects of replicates in Analyze Factorial Design.

The weights are the reciprocal of the fitted variance (1 / fitted variance). Minitab stores weights in every row of your design, even though the standard deviation is missing in some rows. In this case, Minitab uses the same weight at identical combinations of factor settings, unless there are covariates in your model.  

Weights for repeats (adjusted weights)

Store adjusted weights using the adjusted variance, if your data contain repeat measurements with some replicated points. If you do not have replicates, you cannot store adjusted weights from your model. Use the adjusted weights when analyzing the location effects of the stored means of repeat measurements. You must specify these means in DOE > Factorial >Analyze Variability > Storage for Minitab to use them in calculating the adjusted weights.

The weights are estimates of the reciprocal variance of the means. This variance includes both the variance of repeats from your analysis and the variance of the replicates. The adjustment adds in the contribution due to the replicate variance, which is assumed to be constant across factor settings.

If you have covariates in your location model, account for them in the adjusted variance.