remove batch effect

[#1]

I have a data set coming from two independent experiments. I see for some samples there are batch to batch response (final outcome) variation. Please guide me in- how to mitigate this batch effect in JMP.

Thanks 

[#2]

The most obvious approach is to add a data column with the indication of the batch for each row (experimental run) and then include this data column as a covariate in your linear regression model. Be sure to define the modeling type of the batch column as nominal and do not include it in any interaction terms.

[#3]

Mark

Can you elaborate on your last statement?  Why would you not include the batch in any interaction terms?  Absent interactions, aren't you permitting only the intercept of the model to vary by batch, whereas interactions also permit the slope coefficients to vary?

[#4]

I suppose that you could involve batch in interactions. I was stuck in a blocking mindset, I guess, when I wrote that reply. Blocks are seen as homogeneous groups of experimental units, an external source of variation in the response. We don't include blocking factors in interactions, because that would mean that the effects of factors change across blocks. If that were true, then we need to identify the underlying factor and not simply account for a difference in the mean response. We need to clearly distinguish between the meaning and role of concepts like blocking and restricted randomization from the fixed effect of factors of interest.

JMP warns you when you try to include a blocking factor in a model as anything other than a main fixed or random effect.

[#5]

I'm a little confused by your request...There is just not enough context to your request to provide appropriate advice.

Do you want to mitigate (remove the effect of batch) or understand why there is a batch effect?

Is it possible to share the experiments?

If batch variability is noise, there are multiple strategies to handle this.  

1. One would be to use RCBD.  Run one replicate of the experiment with one batch and the other replicate with another batch (confound the block with batch).

2. If some characteristic about the batch is measurable (e.g., viscosity, chemistry, etc), then certainly you could use that measure of the batch as a covariate as Mark suggests.

3. If you want to understand batch variation, you might start with some nested sampling (e.g., measurement system, within batch, batch-to-batch)