Common R Errors
& Model-Fitting Pitfalls

Agenda

Top-10 everyday R errors (from the net)
OLS intuition: why invertibility matters
Errors in lm, aov, lmer, lme
Rank deficiency & dummy-variable trap
Multicollinearity & numerical instability
Over-parameterisation & separation
Debugging toolbox (traceback, browser, rlang)

1 Top-10 Everyday R Errors

(Blog source at the end)

#1 Unmatched brackets / quotes

mean(c(1, 7, 13          # ← missing ')'

Error: unexpected end of input

mean(c(1, 7, 13))        # ✓ fixed

#2 Package / function not loaded

ggplot(mtcars)            # not yet loaded

Error in ggplot(mtcars) : could not find function "ggplot"

install.packages("ggplot2"); library(ggplot2)
ggplot(mtcars) + geom_point(aes(mpg, wt))

#3 Typos in object names

sumary(cars)              # oops

Error in sumary(cars) : could not find function "sumary"

summary(cars)             # ✓

#4 Missing / misspelled arguments

seq(1, 10, lngth = 2)

Error in seq.default(1, 10, lngth = 2) : 
  unused argument (lngth = 2)

seq(1, 10, length = 2)    # ✓

#5 Wrong data type (factor ↔ numeric)

x <- factor(1:3)
mean(x)

Warning in mean.default(x) : argument is not numeric or logical: returning NA
[1] NA

mean(as.numeric(as.character(x)))   # ✓

#6 Forgot `+` in ggplot2

ggplot(mtcars, aes(mpg, wt))
geom_point()

Error: object 'geom_point' not found

ggplot(mtcars, aes(mpg, wt)) +
  geom_point()

#7 Confusing `=` with `==`

if(x = 5) "hi"

Error: unexpected '=' in "if(x ="

if(x == 5) "hi"            # ✓

#8 “Undefined columns selected”

mtcars[, "mpgX"]

Error in `[.data.frame`(mtcars, , "mpgX") : 
  undefined columns selected

mtcars[, "mpg"]            # ✓

#9 File not found / bad path

read.csv("data/myfile.csv")

Error in file(file, "rt") : cannot open the connection
In addition: Warning message:
In file(file, "rt") :
  cannot open file 'data/myfile.csv': No such file or directory

read.csv(file.choose())    # ✓

#10 Misusing `$` on atomic vectors

v <- 1:10
v$foo

Error in v$foo : $ operator is invalid for atomic vectors

list(foo = v)$foo          # ✓

2 OLS Intuition
Why invertibility matters

Visual road-map of β̂ = (X′X)⁻¹X′y

(p×p) ⁻¹ X′y → (X′X)⁻¹ × X′y = β̂

Invertibility 🔑

Singular means some columns are linear combinations of others
If X′X is singular (rank < p) → cannot invert → no unique β̂.
Causes: perfect collinearity, dummy-variable trap, n < p, etc.
R’s lm() cleverly drops redundant columns, but still in left to right order.

3 Errors in `lm`, `aov`, `lmer`, `lme`

Simulated examples & fixes

`lm()` Error 1 – Singular Fit

Message: “coefficients not defined because of singularities”

df <- data.frame(
  y  = rnorm(10),
  x1 = 1:10,
  x2 = 51:60,      # collinear with x1
  x3=31:40
)
lm(y ~ x1 + x2+x3, data = df)

Fix: drop redundant predictor

lm(y ~ x1, data = df)

`lm()` Lengths Differ

Message: “variable lengths differ”

x <- 1:10
y <- rnorm(9)
lm(y ~ x)          # unequal lengths

Error in model.frame.default(formula = y ~ x, drop.unused.levels = TRUE) : 
  variable lengths differ (found for 'x')

Fix: align vector lengths.

`aov()` Grouping Factor Has One Level

df <- subset(PlantGrowth, group == "ctrl")
aov(weight ~ group, data = df)   # fails

Error in `contrasts<-`(`*tmp*`, value = contr.funs[1 + isOF[nn]]) : 
  contrasts can be applied only to factors with 2 or more levels

Fix: ensure ≥ 2 levels.

`lmer()` Error 1 – One-Level Random Effect

Message: “number of levels of a grouping factor is 1”

library(lme4)
df <- data.frame(
  y = rnorm(10),
  x = rnorm(10),
  group = rep("A", 10)
)
lmer(y ~ x + (1 | group), data = df)

Error: grouping factors must have > 1 sampled level

Fix: add more group levels.

`lmer()` Error 2 – Convergence / NaNs

df <- sleepstudy
df$Days <- df$Days * 1000   # poor scaling
lmer(Reaction ~ Days + (Days | Subject), data = df)

 In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv,  :
  Model failed to converge with max|grad| = 3.50551 (tol = 0.002, component 1)
In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv,  :
  Model is nearly unidentifiable: very large eigenvalue
 - Rescale variables?;Model is nearly unidentifiable: large eigenvalue ratio

Fix:

scale/center predictors

check correlation between random intercept and random slope

4 Rank Deficiency & Dummy-Variable Trap

Perfect collinearity demo

set.seed(1)
d <- data.frame(
  y = rnorm(40),
  x1 = rnorm(40)
)
d$x2 <- d$x1        # duplicate
lm(y ~ x1 + x2, d)  # x2 dropped (NA coef)

Dummy-variable trap

iris$Species <- factor(iris$Species)
X.bad <- cbind(1, model.matrix(~ Species + 0, iris))  # intercept + all dummies
qr(X.bad)$rank  # 3 < 4 ⇒ singular

Fix by dropping one level (~ Species) or omitting intercept (~ Species + 0).

5 Multicollinearity & Numerical Instability

Longley example

data(longley)
m <- lm(Employed ~ ., longley)
kappa(m)        # ≫ 10^4 : ill-conditioned

High condition # → unstable estimates.

Floating-point quirks

(1e16 + 1) - 1e16       # 0
sqrt(2)^2 - 2           # 4.4e-16

Rescale predictors; center & scale.
Use ridge / lasso for severe collinearity.

6 Over-Parameterisation & Separation

n <- 5
df <- as.data.frame(matrix(rnorm(n*7), n))
names(df) <- c("y", paste0("x",1:6))
lm(y ~ ., df)  # several NA betas (rank ≤ n)

Need ≥ observations than parameters.

7 Debugging Toolbox

After an error – `traceback()`

f <- function(x) {
  m<-x
  g(x)
  }

g <- function(x) {
  mm=x
  m2=mm
  x + 1}
f("a")              # error
traceback()         # g → f

 traceback() 
2: g(x) at #3
1: f("a")

Interactive – `browser()` / `debug()`

calc <- function(a,b){
  browser()         # pause here
  a/b
}
calc(1,0)

n next, c continue, Q quit.

Tidyverse trace – `rlang`

library(dplyr)
mtcars %>% mutate(z = log(mpgX))  # typo
rlang::last_error()
rlang::last_trace()

Key Take-aways

Check basics first (Top-10 errors) before deep dives.
Invertibility of X′X underpins OLS – watch rank!
Debug disciplined: reproduce → traceback → browser.

Resources

“Top 10 R Errors” – statsandr.com
?lm, ?lmer, ?lme docs & vignettes
R Inferno – P. Burns

Thanks!

Questions?

Common R Errors& Model-Fitting Pitfalls