MatrixEQTL's linearModel proclaims to have nearly identical results to R's lm() function (See www.bios.unc.edu/research/genomic_software/Matrix_eQTL/TestCode3.html).
When I manually calculate my linear model with lm(), I occasionally get inconsistent results. Eventually I noticed that I get the same results if there are no missing values. But with missing values, MatrixEQTL seems to inflate the t-values and thus p-values.
Note: The model in question includes covariates. Additionally, matrixEQTL was conducted on a large dataset. But for curiosity, I started to check individual models and noticed the discrepancy.
Searching online, I found one post that stated "...MatrixEQTL imputes the missing values before doing the calculations." (adairama.wordpress.com/2013/10/25/278/). But I couldn't find other references to imputation. Looking at the manuscript, I get the impression some short-cuts are being made to make these calculations manageable. Perhaps the degrees of freedom aren't being adjusted with missing values. But this is speculation on my part and I would really like to know what is causing the inconsistent results.
Thank you for any help