[R] newdata for predict.lm() ??

[Boris Steipe]

Can't get data from a data frame into predict() without a detour that seems quite unnecessary ...

Reprex:

# Data frame with simulated data in columns "h" (independent) and "w" (dependent)
DAT <- structure(list(h = c(2.174, 2.092, 2.059, 1.952, 2.216, 2.118, 
                            1.755, 2.060, 2.136, 2.126, 1.792, 1.574,
                            2.117, 1.741, 2.295, 1.526, 1.666, 1.581,
                            1.522, 1.995), 
                      w = c(90.552, 89.518, 84.124, 94.685, 94.710, 82.429,
                            87.176, 90.318, 76.873, 84.183, 57.890, 62.005,
                            84.258, 78.317,101.304, 64.982, 71.237, 77.124,
                            65.010, 81.413)),
                 row.names = c( "1",  "2",  "3",  "4",  "5",  "6",  "7", 
                                "8",  "9", "10", "11", "12", "13", "14",
                               "15", "16", "17", "18", "19", "20"),
                 class = "data.frame")


myFit <- lm(DAT$w ~ DAT$h)
coef(myFit)

# (Intercept)       DAT$h 
#   11.76475    35.92002 


# Create 50 x-values with seq() to plot confidence intervals
myNew <- data.frame(seq(min(DAT$h), max(DAT$h), length.out = 50))

pc <- predict(myFit, newdata = myNew, interval = "confidence")

# Warning message:
# 'newdata' had 50 rows but variables found have 20 rows 

# Problem: predict() was not able to take the single column in myNew
# as the independent variable.

# Ugly workaround: but with that everything works as expected.
xx <- DAT$h
yy <- DAT$w
myFit <- lm(yy ~ xx)
coef(myFit)

myNew <- data.frame(seq(min(DAT$h), max(DAT$h), length.out = 50))
colnames(myNew) <- "xx"  # This fixes it!

pc <- predict(myFit, newdata = myNew, interval = "confidence")
str(pc)

# So: specifying the column in newdata to have same name as the coefficient
# name should work, right?
# Back to the original ...

myFit <- lm(DAT$w ~ DAT$h)
colnames(myNew) <- "`DAT$h`"
# ... same error

colnames(myNew) <- "h"
# ... same error again.

Bottom line: how can I properly specify newdata? The documentation is opaque. It seems the algorithm is trying to EXACTLY match the text of the RHS of the formula, which is unlikely to result in a useful column name, unless I assign to an intermediate variable. There must be a better way ...



Thanks!
Boris