ownPersonalSSEsus: Create Personalized F-Statistic

In NickCH-K/MagnifiedIV: Magnified Instrumental Variables Estimator

Description

This function creates an F-statistic of the personalized kind described in Huntington-Klein (2019).

Usage

ownPersonalSSEsus(co, excl, mpred, sse0, n, k)

Arguments

co	This is a vector of coefficients unique to one individual, containing only the coefficients that vary across the sample. For example, if the grouped regression model contains z*group, then co would be a single value containing an individual's z effect.
excl	This is a data frame containing just the variable(s) over which the effects vary. So if the regression model contains z*group, this would be a data frame containing only z.
mpred	This is a vector containing the regression model prediction *if all variables in excl were set to 0*.
sse0, n, k	These are the sum of squared errors, the number of observations, and the number of coefficients in the original regression model.

Details

In particular, it calculates (N-K)*Var(x-hat | individual coefficients)/Var(x-hat).

In Magnified IV this is raised to the power p to be used as a regression weight. This funciton is largely used internally for magnifiedIV() but it is exported here for use in case you'd like to use it to create your own weights. It is perhaps not as user-friendly as it could be, but it is largely an internal function.

Examples

df <- data.frame(w1 = rnorm(1000),
                 w2 = rnorm(1000),
                 e1 = rnorm(1000),
                 e2 = rnorm(1000),
                 z = rnorm(1000),
                 groups = factor(floor(0:999/100)))
df$x <- df$w1+df$w2+df$z+df$e1

fsmodel <- lm(x ~ z*groups + w1 + w2, data = df)

sse0 <- sum((fsmodel$residuals)^2)

indivfx <- factorPull(fsmodel,
                      data = df,
                      factor = 'groups',
                      interaction = 'z',
                      addterm = 'z')

# Create prediction for each observation with just the coefficients that are the
# same for everyone
mpred <- rowSums(t(coef(fsmodel)[c('(Intercept)','w1','w2')]*
                     t(model.matrix(fsmodel)[,c('(Intercept)','w1','w2')])))
# Get a data frame of just the variables with effects that vary
excl <- data.frame(z = df[['z']])

# Get the relevant N and K
n <- nrow(df)
k <- length(fsmodel$coefficients)

# And finally produce that individualized F statistic
indiv.F <- sapply(indivfx, function(x)
  ownPersonalSSEsus(x, excl, mpred, sse0, n, k))

NickCH-K/MagnifiedIV documentation built on Jan. 21, 2020, 11:38 a.m.