Work with categorical variables

Check the role of categorical variables

Frequencies

When you want to check whether, for instance, more male are present in a specific group, you can get a clean contingency table with a chi-square test by following code. Of course, you can work with all the different options as you wish:

library(sjPlot)

tab_xtab(df$VAR1,df$VAR2,
         show.cell.prc = FALSE,
         show.row.prc = TRUE,
         show.col.prc = FALSE,
         show.legend = TRUE,
         show.na = FALSE,
         show.summary=TRUE)

Group	Gender		Total
	Male	Female
Group 1	28 50 %	28 50 %	56 100 %
Group 2	22 45.8 %	26 54.2 %	48 100 %
Total	50 48.1 %	54 51.9 %	104 100 %
χ2=0.052 · df=1 · φ=0.042 · p=0.820

observed values
% within Group

(M)ANOVA

The manovaR function of the CaviR package allows to have a full and informative overview of the role of a categorical variable in prediction of several numeric variables.

The function provides:

The function provides:

• descriptive statistics for each level of the categorical variable

• univariate analyses with a p-value for statistical significance and a partial eta-squared for practical significance

• multivariate analysis with the Wilks’ Lambda

How to interpret the partial eta-squared?

• small when η_p² > .0099

• medium when η_p² > .0588

• large when η_p² >.1379

  (Cohen 2013)

library(CaviR)
manovaR(data[,c('Group','Autonomy','Vitality','Persistence')],
        stand=TRUE, sign = 0.05, tukey = TRUE)

[1] "Grouping variable has only 2 levels. Tukey not applicable"

variables	Group 1	Group 2	F-value	p-value		eta-squared
Autonomy	3.37 (±0.92)	2.31 (±1.09)	57.48	<.001	***	0.37
Vitality	3.37 (±0.92)	2.31 (±1.09)	41.78	<.001	***	0.29
Persistence	3.37 (±0.92)	2.31 (±1.09)	28.18	<.001	***	0.22
Wilks Lambda = 0.607,F(3,95) = 20.479 , p = <.001

When number of levels >2

When the categorical predictor has more than two levels, the function adds the solution of a multicomparison tukey post-hoc analyses to the table in letters. These letters are sorted based on the descriptives.

library(CaviR)
manovaR(data[,c('Groups','Autonomy','Vitality','Persistence')],
        stand=TRUE, sign = 0.05, tukey = TRUE)

variables	Group 1	Group 2	Group 3	Group 4	F-value	p-value		eta-squared
Autonomy	4.31 (±1.19) B	4.23 (±0.95) B	2.73 (±1.46) A	2.16 (±1.16) A	20.18	<.001	***	0.38
Vitality	4.23 (±1.02) B	4.27 (±0.87) B	2.69 (±1.40) A	2.83 (±1.40) A	13.72	<.001	***	0.30
Persistence	3.36 (±1.08) BC	3.38 (±0.78) C	2.64 (±1.15) AB	1.96 (±0.93) A	11.67	<.001	***	0.26
Wilks Lambda = 0.534,F(9,226.488) = 7.396 , p = <.001

References

Cohen, Jacob. 2013. Statistical Power Analysis for the Behavioral Sciences. Academic Press.