R/descrip.md

In haberdashPI/wright.lab.util: A set of utilitiy functions used in our lab.

The key point to remember is that, for a data set with the columns day, sid, pretest, condition and threshold, we want to find:

 threshold - slope.threshold * (pretest - mean(pretest))

The way I've written this is a bit messy because I wasn't as familiar with how to elegantly manipulate data in R when I wrote this code. Let me describe it, step by step.

For this description I'll reference an example using the following call.

covariate.adjust(df,'day','threshold','pretest')

Okay, on to the function definition:

covariate.adjust <- function(df,x,y,covariate,prefix='adj.'){

The first thing we do is set up the name of a variable for the slope we're going to calculate. In the example above, slope.name = 'slope.threshold'.

  slope.name = paste(prefix,'slope.',y,sep='')

We then estimate a separate slope for the variables threshold (y) and pretest (covariate) for each day (x). For instnace, for the post-test day, this is just the slope betwene the pre- and post-test (which is what we normally plot in our pre- vs. post-test plots). This is done using ddply which applies the given function to each subset of df with a different value for x.

  slopes = ddply(df,x,function(df){

The first step to find the slope is to perform a regression using lm. The below call to lm would be equivalent, in the working example, to lm(threshold ~ pretest). We then use coef(...bla bla...)[2] to extract the second coefficient of the regression (the slope).

    df = data.frame(slope=
        coef(lm(reformulate(response = y, termlabels=covariate),df))[2])

Then we rename the column so it matches the slope.name we defined above ("slope.threshold").

    colnames(df) = slope.name
    df
  })

After calculating the slopes, we add the slopes to the data using merge. Merge combines two data frames by matching the columns that are shared, and adding columns that are not, repeating them if the shared columns are the same. Since we calculated a different slope for each day, merging these slopes with df will add a new column (slope.threshold) that will be the same for all rows with the same day (but different for rows with different days).

  df = merge(df,slopes)

Now that we've calculated a slope between threshold and pre-test for each day we can correct for the pre-test. First we create a new column called 'mean.pretest' (paste('mean',covariate,sep='.')) and set it equal to the mean of the pretest.

  df[,paste('mean',covariate,sep='.')] = mean(df[,covariate],na.rm=TRUE)

Then, finally, we correct for the pre-test. The below code is equivalent to the description I provided at beginning for the example.

  df[,paste(prefix,y,sep='')] = df[,y] - df[,slope.name] *
    (df[,covariate] - mean(df[,covariate],na.rm=TRUE))
  df
}

haberdashPI/wright.lab.util documentation built on Nov. 4, 2019, 1:25 p.m.