nes_1976 <- nes %>%
  filter(year == 1976)
nes_1976$rep_pres_intent <- as.factor(nes_1976$rep_pres_intent)
myvars <- c("ideo7", "black", "female", "rep_pres_intent", "presapprov", "age",
            "educ1", "urban", "income", "union", "perfin1")

newdata <- nes_1976[myvars]

data.num <- newdata %>%
  mutate(ideo7=as.numeric(ideo7)) %>%
  mutate(black=as.numeric(black)) %>%
  mutate(female=as.numeric(female)) %>%
  mutate(presapprov=as.numeric(presapprov)) %>%
  mutate(educ1=as.numeric(educ1)) %>%
  mutate(urban=as.numeric(urban)) %>%
  mutate(income=as.numeric(income)) %>%
  mutate(union=as.numeric(union)) %>%
  mutate(perfin1=as.numeric(perfin1))

lm2 <- glm(rep_pres_intent ~., data = data.num, family = binomial)
summary(lm2)

mmplot(lm2)

## this blows up the memory allocated to it.  Less smooth fits require
## more memory. After looking at help(mmplot), help(locfit),
## help(locfit.raw), and help(lp) [and some web spelunking], I
## determined that the following makes a reasonable tradeoff between
## not blowing up memory and not having too smooth a fit:
##
##
##   mmplot(lm2,locfit.control=list(nn=1),layout=c(4,3))
##
##
## Details on the parameters that can affect smoothing (with different
## names) can be found in Chapter 2 of
##
## Loader, C. (1999). Local regression and likelihood. Springer
## Science & Business Media.
##
## A pdf of the book can be had for free from springer link.