The oneway function wraps a number of analysis of variance functions into one convenient interface that is similar to the oneway anova command in SPSS.
oneway(
y,
x,
posthoc = NULL,
means = FALSE,
fullDescribe = FALSE,
levene = FALSE,
plot = FALSE,
digits = 2,
omegasq = TRUE,
etasq = TRUE,
corrections = FALSE,
pvalueDigits = 3,
t = FALSE,
conf.level = 0.95,
posthocLetters = FALSE,
posthocLetterAlpha = 0.05,
overrideVarNames = NULL,
silent = FALSE
)
# S3 method for oneway
print(
x,
digits = x$input$digits,
pvalueDigits = x$input$pvalueDigits,
na.print = "",
...
)
# S3 method for oneway
pander(
x,
digits = x$input$digits,
pvalueDigits = x$input$pvalueDigits,
headerStyle = "**",
na.print = "",
...
)y has to be a numeric vector.
x has to be vector that either is a factor or can be converted into one.
Which post-hoc tests to conduct. Valid values are any correction methods in p.adjust.methods (at the time of writing of this document, "holm", "hochberg", "hommel", "bonferroni", "BH", "BY", "fdr", "none"), as well as "tukey" and "games-howell".
Whether to show the means for the y variable in each of the groups determined by the x variable.
If TRUE, not only the means are shown, but all statistics acquired through the 'describe' function in the 'psych' package are shown.
Whether to show Levene's test for equality of variances (using
car's leveneTest function but specifying
mean as function to compute the center of each group).
Whether to show a plot of the means of the y variable in each of the groups determined by the x variable.
The number of digits to show in the output.
Whether to show the omega squared effect size.
Whether to show the eta squared effect size (this is biased and generally advised against; omega squared is less biased).
Whether to show the corrections for unequal variances (Welch and Brown-Forsythe).
The number of digits to show for p-values; smaller p-values will be shown as <.001 or <.0001 etc.
Whether to transpose the dataframes with the means (if requested) and the anova results. This can be useful for blind people.
Confidence level to use when computing the confidence interval for eta^2. Note that the function we use doubles the 'unconfidence' level to maintain consistency with the NHST value (see http://yatani.jp/HCIstats/ANOVA#RCodeOneWay, http://daniellakens.blogspot.nl/2014/06/calculating-confidence-intervals-for.html or Steiger, J. H. (2004). Beyond the F test: Effect size confidence intervals and tests of close fit in the analysis of variance and contrast analysis. Psychological methods, 9(2), 164-82. doi:10.1037/1082-989X.9.2.164
Whether to also compute and show the letters
signifying differences between groups when conducting post hoc tests. This
requires package multcompView to be installed.
The alpha to use when determining whether groups
have different means when using posthocLetters.
Can be used to override the variable names (most useful in functions).
Whether to show warnings and other diagnostic information or remain silent.
How to print missing values.
Any additional arguments are passed to the print or
pander function.
The header pre- and suffix to use when pandering the result (useful when working with Markdown).
A list of three elements:
List with input arguments
List of intermediate objects, such as the aov and Anova (from the car package) objects.
List with etasq, the effect size, and dat, a dataframe with the Oneway Anova results.
By my knowledge the Brown-Forsythe correction was not yet available in R. I took this from the original paper (directed there by Field, 2014). Note that this is the corrected F value, not the Brown-Forsythe test for normality!
Brown, M., & Forsythe, A. (1974). The small sample behavior of some statistics which test the equality of several means. Technometrics, 16(1), 129-132. https://doi.org/10.2307/1267501
Field, A. (2014) Discovering statistics using SPSS (4th ed.). London: Sage.
Steiger, J. H. (2004). Beyond the F test: Effect size confidence intervals and tests of close fit in the analysis of variance and contrast analysis. Psychological methods, 9(2), 164-82. doi:10.1037/1082-989X.9.2.164
### Do a oneway Anova
oneway(y=ChickWeight$weight, x=ChickWeight$Diet);
#> ### Oneway Anova for y=weight and x=Diet (groups: 1, 2, 3, 4)
#>
#> Omega squared: 95% CI = [.02; .09], point estimate = .05
#> Eta Squared: 95% CI = [.02; .08], point estimate = .05
#>
#> SS Df MS F p
#> Between groups (error + effect) 155862.66 3 51954.22 10.81 <.001
#> Within groups (error only) 2758693.27 574 4806.09
#>
### Also order means and transpose the results
oneway(y=ChickWeight$weight, x=ChickWeight$Diet, means=TRUE, t=TRUE);
#> ### Oneway Anova for y=weight and x=Diet (groups: 1, 2, 3, 4)
#>
#> Omega squared: 95% CI = [.02; .09], point estimate = .05
#> Eta Squared: 95% CI = [.02; .08], point estimate = .05
#>
#> Between groups (error + effect) Within groups (error only)
#> SS 155862.66 2758693.27
#> Df 3 574
#> MS 51954.22 4806.09
#> F 10.81
#> p <.001
#>
#> ### Means for y (weight) separate for each level of x (Diet):
#>
#> Diet = 1:
#>
#> n 220.0
#> mean 102.6
#> sd 56.7
#> median 88.0
#> se 3.8
#>
#> Diet = 2:
#>
#> n 120.0
#> mean 122.6
#> sd 71.6
#> median 104.5
#> se 6.5
#>
#> Diet = 3:
#>
#> n 120.0
#> mean 142.9
#> sd 86.5
#> median 125.5
#> se 7.9
#>
#> Diet = 4:
#>
#> n 118.0
#> mean 135.3
#> sd 68.8
#> median 129.5
#> se 6.3