You are on page 1of 6

SESM002

Research Methods & Statistics Seminars:


Analysis of Variance (ANOVA) 2

1-Way ANOVA (Independent Measures)

Maximal oxygen uptake (VO2max) was determined in three groups of individuals: (i)
professional footballers (ii) national standard distance runners (iii) untrained. The data can
be downloaded on the following link, on the “Independent ANOVA” tab:

https://drive.google.com/file/d/15wz--d1Hsu_Msbbs7R7-UzNsxrYmwZoX/view?usp=sharing

1. Insert the data into SPSS and code the three groups as appropriate
2. Produce descriptive statistics and examine the data for normality, skew, kurtosis and
homogeneity of variance
3. Undertake a 1-Way ANOVA to examine for differences between groups
a. Select Analyse → Compare Means → 1 Way ANOVA
b. Place your grouping variable into the “Factor” box and your outcome (VO2max)
into the “Dependent List” (figure 2)
c. Select “Options” and check the Brown-Forsythe and Welch test check boxes;
these will produce statistics that are corrected for a violation of homogeneity of
variance (you will notice that this box also has options for descriptive statistics
and a homogeneity of variance test), see figure 3
i. The Means plot is useful to check if you have not already plotted your
data and “eyeballed” it

Figure 2 Figure 3

d. Select “Post-Hoc” and choose your tests by selecting the relevant boxes (figure 4)
i. Bonferroni will most effectively control the type 1 error rate, but lacks
power as the number of comparisons rises; use Tukey in such cases
ii. REGWQ is a good all rounder unless you have unequal group sizes
iii. If you are concerned about homogeneity of variance, Games Howell is
worthwhile unless you have small sample sizes
1
Figure 5

Figure 4

Figure 6

5. Interpreting the output from a 1-Way ANOVA


a. The “ANOVA” box will provide the headline / omnibus result of the ANOVA, in the
“Sig.” column and the “Between Groups” row (figure 5); you should decide
whether a significant difference between groups is indicated
b. Look also at “Robust Tests of Equality of Means” (figure 6) which provides the
Welch and Brown-Forsythe tests which have corrected the degrees of freedom to
account for any violation of the assumption of homogeneity of variance; does the
output here change your mind?
c. Notice that you can utilise the Welch and Brown-Forsythe tests routinely, rather
than only when a formal test (such as Levene’s test) is significant
d. The “Means Plot” provides a rudimentary figure with which to help interpret the
data which you may wish to consider before you examine “Post-Hoc Tests” (figure
7). The first box (“Multiple Comparisons”) provides a series of corrected t-tests
that were requested via the Analyse menus; you can see that SPSS likes to repeat
itself.
i. Where is it indicated that specific group differences exist?
ii. Do the different post-hoc procedures agree?
e. Some of the Post-Hoc procedures (e.g. REGWQ) produce a “Homogenous
Subset” analysis where groups that are not significantly different from each
other are sequentially paired together. In other words, pairs of groups within a
column on this output are NOT significantly different; pairs of groups that are
not here ARE significantly different.

2
Figure 7
1-Way ANOVA (Repeated Measures)

Peak leg extension force (expressed as a percentage of resting) was measured every hour
during 3-hours of heavy running exercise at 70%VO2max in a group of club athletes. The data
can downloaded on the following link (“Repeated ANOVA” tab):

https://drive.google.com/file/d/15wz--d1Hsu_Msbbs7R7-UzNsxrYmwZoX/view?usp=sharing

1. Insert the data into SPSS and use the variable view to ensure data are correctly titled, the
correct type of data inserted and unnecessary decimal places removed (**shudders**)
2. Produce descriptive statistics and examine the data for normality, skew and kurtosis;
homogeneity of variance is not relevant for a repeated measures design
3. The Repeated Measures ANOVA
a. Select Analyse → General Linear Model → Repeated Measures
b. In the “Within-Subject Factor Name” box, give your factor a name (e.g. “time”)
and assign the correct number of levels (in this case three: 60-mins, 120-mins,
180-mins); click “Add” to enter it into the box and then “Define” to enter the next
step (see figure 9)
c. On the next box (figure 10) use the blue arrow to move the variable names into
the “Within-Subjects Variables” box; the title of this box will also include the
name you gave to the factor (e.g. “time”). Ensure you put the variables in a
worthwhile order with some sort baseline condition as either the first or last
condition

Figure 10

Figure 9

3
d. The “Options” button will give you options for descriptive and further statistics,
such as a test for homogeneity
e. I would be minded not to try and plot a graph in SPSS, and the “Post-Hoc” menu
is all greyed out because of issues with post-hoc testing when data is not
spherical
f. The “EM Means” menu gives some limited options for comparing within your
main effects (of which you have only one in this case)
i. Select this menu and use the blue arrow to move your main effect into
the “Display Means For” box. Check the “Compare Main Effects” box and
choose from LSD (no corrections made for multiple comparisons),
Bonferroni or Sidak (maintains statistical power better when the number
of comparisons are high). In this worksheet I’ve gone for Bonferroni
because there are only 3 conditions (see figure 11).
g. The “Contrasts” menu allows you to select in advance the comparisons you plan
to make (“Planned Comparisons”) and are more powerful because they are not
corrected in the way that Post-Hoc tests are (or should be). I’ve selected
“Repeated” because I am interested in the change in our parameter of interest
(% of peak force) with time (our factor or independent variable); this option will
compare each condition to the previous (see figure 12).
i. You need to highlight your factor, choose the contrast from the
dropdown list and ensure you click “Change”.
ii. If you wanted to compare each condition to the baseline condition, you
could choose “Simple” and then specify the Last or First button,
depending on how you set your data up (from figure 10).

Figure 11 Figure 12

4. Run the analysis by clicking “OK” on the “Repeated Measures” box (figure 10) and observe
the output
a. One of the assumptions of the ANOVA is sphericity (variances of differences
between groups should be similar). Hence, our first stop is “Mauchly’s Test of
Sphericity” (figure 13). The significance level here will tell you if this test is failed
(i.e. p<0.05).
b. If this is failed, we should take forward either the Greenhouse-Geisser or Huynh-
Feldt correction. As a rule of thumb, if the Epsilon of GG is <0.75, use GG. If the4
Epsilon of GG is >0.75, use HF.
Figure 13

c. Decide how to proceed following the test of sphericity and then move to the main
output of the ANOVA: “Tests of Within Subject Effects” (figure 14). The significance
level (p-value) is in the final column of this table. The four values relate to either
sphericity assumed, or one of the corrections to be made if sphericity cannot be
assumed. You should now be able to assign a p-value to the test of difference
(ANOVA) you have just undertaken.
d. Field (2018) suggests that sphericity should never be assumed, even if Mauchly’s
Test is non-significant. Rather, based on the previous “rules of thumb”, a correction
should always be applied to your ANOVA and the p-value reported accordingly.

Figure 14

e. Assuming a significant difference is present, you can therefore move to either your
planned contrasts or post-hoc comparisons. You should really only select for one of
these, but we’ll look at both for completeness
i. Figure 15 shows the output of the repeated contrasts that were requested.
You can see that sequential comparisons are produced (Level 1 vs Level 2,
Level 2 vs Level 3) which are often a useful analysis for data produced with
time. Notice that this analysis does not produce a Level 1 vs Level 3
comparison, but you should have decided in advance that this was not
necessary. The final (Sig.) column gives the level of significance; you should
decide which time points were significantly different to the others. 5
Figure 15

i. The post-hoc tests (with any correction that was selected for) can be found
in the “Pairwise Comparisons” table (figure 16) under “Estimated Marginal
Means”, a section that also includes some useful descriptive statistics. As
ever, SPSS repeats itself but returns ALL possible comparisons. You should
determine where the significant differences between conditions are.

Figure 16

5. You should practice writing out some statements that would appear in your results section
on the back of the data you have analysed. Journal articles are a good place to look with
examples here: https://physoc.onlinelibrary.wiley.com/doi/full/10.1113/EP086304 though
you may be more formal and include the t-score and degrees of freedom as indicated in the
lecture content
6. If you are finished you should use SPSS or Excel to draw a high quality figure of each of your
data outputs in a style you would associate with a journal. This figure should include all
required elements (such as a legend) and the insertion of symbols to indicate the specific6
location of significant effects.

You might also like