4

I would like to do Tukey HSD post hoc tests for a repeated measure ANOVA. The entered formula "TukeyHSD" returns me an error. I can't find the answer in the forum. Can I ask for help?

"treat" is repeated measures factor, "vo2" is dependent variable.

Below is a script that is producing this error:

my_data <- data.frame(
  stringsAsFactors = FALSE,
  id = c(1L,2L,3L,4L, 5L,1L,2L,3L,4L,5L,1L,2L,3L,4L,5L,1L,2L,3L,4L,5L),
  treat = c("o","o","o","o","o","j","j","j","j","j","z","z","z","z","z","w","w","w","w","w"),
  vo2 = c("47.48","42.74","45.23","51.65","49.11","51.00","43.82","49.88","54.61","52.20","51.31",
          "47.56","50.69","54.88","55.01","51.89","46.10","50.98","53.62","52.77"))

summary(rm_result <- aov(vo2~factor(treat)+Error(factor(id)), data = my_data))
TukeyHSD(rm_result, "treat", ordered = TRUE)
alunats
  • 89
  • 5

2 Answers2

2

TukeyHSD() can't work with the aovlist result of a repeated measures ANOVA. As an alternative, you can fit an equivalent mixed effects model with e.g. lme4::lmer() and do the post-hoc tests with multcomp::glht().

my_data$vo2 <- as.numeric(my_data$vo2)
my_data$treat <- factor(my_data$treat)
m <- lme4::lmer(vo2 ~ treat + (1|id), data = my_data)
summary(multcomp::glht(m, linfct=mcp(treat="Tukey")))

# Simultaneous Tests for General Linear Hypotheses
# 
# Multiple Comparisons of Means: Tukey Contrasts
# 
# 
# Fit: lmer(formula = vo2 ~ treat + (1 | id), data = my_data)
# 
# Linear Hypotheses:
#            Estimate Std. Error z value Pr(>|z|)    
# o - j == 0   -3.060      0.583  -5.248   <0.001 ***
# w - j == 0    0.770      0.583   1.321   0.5497    
# z - j == 0    1.588      0.583   2.724   0.0327 *  
# w - o == 0    3.830      0.583   6.569   <0.001 ***
# z - o == 0    4.648      0.583   7.972   <0.001 ***
# z - w == 0    0.818      0.583   1.403   0.4974    
# ---
# Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
# (Adjusted p values reported -- single-step method)

Comparison of the mixed effects model's ANOVA table with your repeated measures ANOVA results shows that both approaches are equivalent in how they treat the treat variable:

anova(m)
# Analysis of Variance Table
#       npar Sum Sq Mean Sq F value
# treat    3 61.775  20.592   24.23

summary(rm_result)
# Error: factor(id)
#           Df Sum Sq Mean Sq F value Pr(>F)
# Residuals  4  175.9   43.98               
# 
# Error: Within
#               Df Sum Sq Mean Sq F value   Pr(>F)    
# factor(treat)  3  61.78   20.59   24.23 2.22e-05 ***
# Residuals     12  10.20    0.85                     
# ---
# Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Robert Hacken
  • 3,878
  • 1
  • 13
  • 15
1

Alternatively, you could also do it as in the reprex below. Note that the cld() part is optional and simply tries to summarize the results via the "Compact Letter Display" (details on it here)

# data --------------------------------------------------------------------
my_data <- data.frame(
  stringsAsFactors = FALSE,
  id = c(1L,2L,3L,4L, 5L,1L,2L,3L,4L,5L,1L,2L,3L,4L,5L,1L,2L,3L,4L,5L),
  treat = c("o","o","o","o","o","j","j","j","j","j","z","z","z","z","z","w","w","w","w","w"),
  vo2 = c("47.48","42.74","45.23","51.65","49.11","51.00","43.82","49.88","54.61","52.20","51.31",
          "47.56","50.69","54.88","55.01","51.89","46.10","50.98","53.62","52.77"))

my_data$vo2 <- as.numeric(my_data$vo2)
my_data$treat <- factor(my_data$treat)


# model -------------------------------------------------------------------
m <- lme4::lmer(vo2 ~ treat + (1|id), data = my_data)


# emmeans -----------------------------------------------------------------
library(emmeans)
emmeans <- emmeans(m, specs = "treat")
pairs(emmeans, adjust = "Tukey")
#>  contrast estimate    SE df t.ratio p.value
#>  j - o       3.060 0.583 12   5.248  0.0010
#>  j - w      -0.770 0.583 12  -1.321  0.5681
#>  j - z      -1.588 0.583 12  -2.724  0.0761
#>  o - w      -3.830 0.583 12  -6.569  0.0001
#>  o - z      -4.648 0.583 12  -7.972  <.0001
#>  w - z      -0.818 0.583 12  -1.403  0.5209
#> 
#> Degrees-of-freedom method: kenward-roger 
#> P value adjustment: tukey method for comparing a family of 4 estimates


# multcomp ----------------------------------------------------------------
library(multcomp)
library(multcompView)
cld(emmeans, Letters = letters)
#>  treat emmean   SE   df lower.CL upper.CL .group
#>  o       47.2 1.53 4.47     43.2     51.3  a    
#>  j       50.3 1.53 4.47     46.2     54.4   b   
#>  w       51.1 1.53 4.47     47.0     55.1   b   
#>  z       51.9 1.53 4.47     47.8     56.0   b   
#> 
#> Degrees-of-freedom method: kenward-roger 
#> Confidence level used: 0.95 
#> P value adjustment: tukey method for comparing a family of 4 estimates 
#> significance level used: alpha = 0.05 
#> NOTE: If two or more means share the same grouping symbol,
#>       then we cannot show them to be different.
#>       But we also did not show them to be the same.

Created on 2022-12-20 with reprex v2.0.2

Paul Schmidt
  • 1,072
  • 10
  • 23