1
$\begingroup$

I can't get the same results in R as in GraphPad Prism for repeated measures anova.

The experiment was a stimulation time course, so I have as DV=response and as factor "time" within groups, also I add a factor sample for each experiment

data <- read.csv("http://dl.dropbox.com/u/4828275/datos.csv")
options(contrasts=c("contr.sum","contr.poly"))

## Convert variables to factor
data <- within(data, {
sample <- factor(sample)
time <- factor(time)
})
aov <- aov(response~time+sample, data=data)
summary(glht(aov, linfct=mcp(time="Dunnett")))

     Simultaneous Tests for General Linear Hypotheses

Multiple Comparisons of Means: Dunnett Contrasts


Fit: lme.formula(fixed = response ~ time, data = data, random = ~1 | 
sample)

Linear Hypotheses:
            Estimate Std. Error z value Pr(>|z|)
2 - 0 == 0    1.1789     2.0800   0.567        1
5 - 0 == 0    1.2966     2.0800   0.623        1
10 - 0 == 0   1.0555     2.0800   0.507        1
15 - 0 == 0   0.4317     2.0800   0.208        1
30 - 0 == 0   0.2148     2.0800   0.103        1
(Adjusted p values reported -- bonferroni method)

For repeated measures I have this code

aov.repeated <- ezANOVA(
  data
  , dv = .(response)
  , wid = .(time)
  , within = .(sample)
  , type = 1
  , return_aov = TRUE
)$aov

The GraphPad Prism results for the same data was

Table Analyzed  Data 1              

Repeated Measures ANOVA                 
  P value   0.0415              
  P value summary   *               
  Are means signif. different? (P < 0.05)   Yes             
  Number of groups  6               
  F 2.863               
  R square  0.4172              

 Was the pairing significantly effective?                   
  R square  0.1980              
  F 2.119               
  P value   0.1162              
  P value summary   ns              
  Is there significant matching? (P < 0.05) No              

ANOVA Table SS  df  MS      
  Treatment (between columns)   130.6   5   26.12       
  Individual (between rows) 77.30   4   19.32       
  Residual (random) 182.4   20  9.121       
  Total 390.3   29          

Dunnett's Multiple Comparison Test  Mean Diff.  q   Significant? P < 0.05?  Summary 95% CI of diff
  0 vs 2    -2.861  1.498   No  ns  -8.085 to 2.362
  0 vs 5    -5.777  3.024   Yes *   -11.00 to -0.5531
  0 vs 10   -6.009  3.146   Yes *   -11.23 to -0.7855
  0 vs 15   -4.621  2.419   No  ns  -9.844 to 0.6029
  0 vs 30   -2.581  1.351   No  ns  -7.805 to 2.642

How can I get the same results as above in R? Is there a way to get Dunnett's Multiple Comparison Test in aov.repeated?

$\endgroup$
3
$\begingroup$

All I have to go by are the labels in your .csv file, but it looks to me like you set the problem up incorrectly in Prism. I transposed your data so each row in Prism is one matched sample. So the data entry looks like this:

enter image description here

Now the results (from GraphPad Prism 5.04) match the results you showed from R. The differences between means match, and the q values in Prism match the z values in R:

enter image description here

The problem is you had told Prism, essentially, that all the values collected at one time point were matched. By transposing, I am telling Prism that all the values from one sample (at multiple time points) were matched. If you choose one-way ANOVA in Prism, and specify repeated measures, it assumes that all values in one row are matched (not that all values in one column are matched).

Download the Prism file.

$\endgroup$
  • $\begingroup$ Thank you so much, I've got an error when importing data into R corrected with time <- c(rep(0,5), rep(2,5), rep(5,5), rep(10,5), rep(15,5), rep(30,5)); sample <- c(rep(1:5, 6)) so the R transposed table was bad; now I've got the same results. $\endgroup$ – friveroll May 28 '12 at 19:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.