EPS 625 – INTERMEDIATE STATISTICS

ONE-WAY ANOVA – IN-CLASS EXAMPLE

ADJUSTING FOR HETEROGENEITY OF VARIANCE
You are interested in whether male college students with black, blond, brunette, or red hair differ
with respect to their social extrovertedness. For this study, 60 male college students from a local
college (15 for each hair color) have been randomly selected to participate in the study using a
stratified random sampling approach. The students were given a measure of social extroversion
with a range of 0 (low level of social extroversion) to 10 (high level of social extroversion).
Conduct an ANOVA to investigate the relationship between hair color and social extroversion.
For this example, use an a priori alpha level of significance of = .05 for each statistical
analysis (except, use an α = .001 for the Shapiro-Wilks test) to answer the following questions.

1. What would the null hypothesis be for this study? Show/write the appropriate symbols or
expression in words.

H0: µblack = µblond = µbrunette = µredhead

The group means (average social extrovertedness scores) for the four (hair color)
groups are equal.

2. What would the research/alternative hypothesis be for this study? Show/write the appropriate
symbols or expression in words.

Ha: µi ≠ µk for some i, k

At least two of the (hair color) group means are significantly different from each
other.

3. Prior to examining whether group means differ, you need to test the assumptions underlying
the one-way ANOVA.
a. Was the assumption of independence met for these data? Indicate how you made this
determination.

YES, the assumption of independence was met. This is indicated by the four
groups of male college students being independent of each other. A stratified
random sampling technique was used, which focused on obtaining a random
selection of male college students for each of the four hair colors.

b. Was the assumption of normality met for these data? Indicate how you made this
determination.

YES, the assumption of normality was met for this set of data. This is indicated
by the fact that none of the standardized skewness values exceeded +3.29, nor
were any of the probability values less than (or equal to) the .001 alpha level set
for the Shapiro-Wilks test.
 − .122 − .050
Black: = −.2103 Blond: = −.0862
.580 .580
.024 .586
Brunette: = .0414 Redhead: = 1.0103
.580 .580

Black: p (.160) > (.001) Blond: p (.402) > (.001)

Brunette: p (.579) > (.001) Redhead: p (.463) > (.001)

c. Was the assumption of homogeneity of variance met for these data? Indicate how you
made this determination.

NO, the assumption of homogeneity was not met, because p (.038) < (.05).
This is indicated by the Levene’s Test of Homogeneity of Variances, F(3, 56) =
3.01, p = .038. With an alpha level of .05, p (.038) < α (.05), which indicates
significance, the null hypothesis (no variance difference) is rejected – as such,
indicating that the assumption of homogeneity of variance is not met.

Test of Homogeneity of Variances

Social Extroversion
Levene
Statistic df1 df2 Sig.
3.007 3 56 .038

Looking at the Variance Ratio, we find that the largest group variance ( σ 2Blond =
2.8752 = 8.2656) is not more than 4-5 times that of the smallest group variance
( σ 2Brunette = 1.4472 = 2.0938). That is, 8.27/2.09 = 3.96.

4. The next question that needs to be answered is whether all of the groups are the same on their
social extroversion means. This is answered by conducting a One-way Analysis of Variance
using hair color as the independent variable and students’ scores on a measure of social
extroversion as the dependent variable. If applicable, use the Welch statistic. What is your
conclusion (at this point) from this analysis? Indicate how you came to your conclusion.

Since the assumption of homogeneity of variance was not met for this data, we used
the obtained Welch’s adjusted F ratio (10.95), which was significant at the .05 alpha
level (p shown as .000, that is p < .001) reported as Welch’s F(3, 30.17) = 10.95, p <
.001 (or, p < .05), we can conclude that at least two of the four hair color groups
differ significantly on their average social extrovertedness scores.

However, beyond that, post hoc follow-up procedures (e.g., Games-Howell) will
need to be conducted to test the difference between all unique pairwise
comparisons.

ADJUSTED ANOVA – IN-CLASS EXAMPLE

PAGE 2
 Robust Tests of Equality of Means

Social Extroversion
a
Statistic df1 df2 Sig.
Welch 10.946 3 30.171 .000
a. Asymptotically F distributed.

5. Calculate the measure of association and interpret its meaning if applicable, or indicate why
this measure would not be needed.

Since we used the Welch’s F test, we will use an adjusted omega squared formula.

df bet ( F − 1) 3(10.946 − 1) 3(9.946) 29.838 29.838

est. ω 2 = = = = = = .3321 = .33
df bet ( F − 1) + NT 3(10.946 − 1) + 60 3(9.946) + 60 29.838 + 60 89.838

Therefore, we conclude that approximately 33% (ω2 = .33) of the total variance in the
dependent variable (student’s average social extrovertedness scores) is accounted
for by the independent variable (hair color with four levels).

6. Write the statistical strand for this one-way ANOVA analysis.

Welch’s F(3, 30.17) = 10.95, p < .001, est. ω2 = .33

7. Assuming that you found a significant F, how do the pairs of groups differ? Indicate which
post hoc procedure you used and why. Indicate your findings from the post hoc analysis.
That is, how did you determine the pair to be significant or not (this must go beyond the * as
an indication)? Be sure to discuss all unique pairwise comparisons.

Games-Howell post hoc procedure is used since the homogeneity of variance

assumption was not met…

Using an a priori alpha level of .05

Black vs. Blond (Mean difference = 1.200) is not significant, p (.563) > α (.05)
Black vs. Brunette (Mean difference = 3.667) is significant, p (.000) < α (.05)

Black vs. Redhead (Mean difference = 2.333) is significant, p (.034) < α (.05)

Blond vs. Brunette (Mean difference = 2.467) is significant, p (.035) < α (.05)
Blond vs. Redhead (Mean difference = 1.133) is not significant, p (.641) > α (.05)
Brunette vs. Redhead (Mean difference = 1.333) is not significant, p (.262) > α (.05)

ADJUSTED ANOVA – IN-CLASS EXAMPLE

PAGE 3
8. For all significant pairwise comparisons, calculate and report the effect size.

Descriptives

Social Extroversion
95% Confidence Interval for
Mean
N Mean Std. Deviation Std. Error Lower Bound Upper Bound Minimum Maximum
Black 15 7.33 2.059 .532 6.19 8.47 4 10
Blond 15 6.13 2.875 .742 4.54 7.73 1 10
Brunette 15 3.67 1.447 .374 2.87 4.47 1 6
Redhead 15 5.00 2.330 .602 3.71 6.29 2 10
Total 60 5.53 2.574 .332 4.87 6.20 1 10

(Std. Deviation)2 = σ2

M1 − M 2 σ 12 + σ 22
Using Cohen’s d d= where σ pooled =
σ pooled 2

4.2395 + 2.0938 6.3333

ES for Black vs. Brunette = σ pooled = = = 3.1667 = 1.7795
2 2
M1 − M 2 7.33 − 3.67 3.667
d= = = = 2.0607 = 2.06
σ pooled 1.7795 1.7795

4.2395 + 5.4289 9.6684

ES for Black vs. Redhead = σ pooled = = = 4.8342 = 2.1987
2 2
M1 − M 2 7.33 − 5.00 2.333
d= = = = 1.0611 = 1.06
σ pooled 2.1987 2.1987

8.2656 + 2.0938 10.3594

ES for Blond vs. Brunette = σ pooled = = = 5.1797 = 2.2759
2 2
M1 − M 2 6.13 − 3.67 2.467
d= = = = 1.0840 = 1.08
σ pooled 2.2759 2.2759

ADJUSTED ANOVA – IN-CLASS EXAMPLE

PAGE 4
 Results

A One-way Analysis of Variance (ANOVA) was used to examine the question of

whether male college students with black, blond, brunette, or red hair differ with respect to their

social extrovertedness. The independent variable represented the different hair colors with four

groups being represented: 1) black; 2) blond; 3) brunette; and 4) red. The dependent variable was

the average score that students made on a measure of social extroversion with a range of 0 (low

level of social extroversion) to 10 (high level of social extroversion). See Table 1 for the means

and standard deviations for each of the four groups.

Table 1

Means and Standard Deviations of Social Extroversion Scores by Hair Color

Hair Color n Mean SD

Black 15 7.33 2.06

Blond 15 6.13 2.88

Brunette 15 3.67 1.45

Red 15 5.00 2.33

Total 60 5.53 2.57

The test for normality, examining standardized skewness and the Shapiro-Wilks test,

indicated the data were statistically normal. However, the Levene’s F test revealed that the

homogeneity of variance assumption was not met (p = .038). As such, the Welch’s F test was

used. An alpha level of .05 was used for all subsequent analyses. The one-way ANOVA of

student’s average score on the measure of social extroversion revealed a statistically significant

ADJUSTED ANOVA – IN-CLASS EXAMPLE

PAGE 5
main effect, Welch’s F(3, 30.17) = 10.95, p < .001, indicating that not all hair colors had the

same average score on the measure of social extroversion. The estimated omega squared (ω2 =

.33) indicated that approximately 33% of the total variation in average score on students’

measure of social extroversion is attributable to differences between the four colors of hair.

Post hoc comparisons, using the Games-Howell post hoc procedure, were conducted to

determine which pairs of the four hair color means differed significantly. These results are given

in Table 2 and indicate that students with black hair (M = 7.33, SD = 2.06) had a significantly

higher average score on the measure of social extroversion than students with brunette hair (M =

3.67, SD = 1.45) as well as students with red hair (M = 5.00, SD = 2.33). The effect sizes for

these two significant effects were 2.06 and 1.06, respectively. Additionally, students with blond

hair (M = 6.13, SD = 2.88) had a significantly higher average score on the measure of social

extroversion than students with brunette hair, with an effect size of 1.08.

Table 2

Post Hoc Results for Social Extroversion Scores by Hair Color

Hair Color Mean Mean Differences ( X i − X j )

(Effect Sizes are indicated in parentheses)

1 2 3 4

1. Black 7.33 −−

2. Blond 6.13 1.20 −−

3. Brunette 3.67 3.66*** 2.46* −−

(2.06) (1.08)

4. Red 5.00 2.33* 1.13 1.33 −−

(1.06)

*p < .05, *** p < .001

ADJUSTED ANOVA – IN-CLASS EXAMPLE

PAGE 6