Violation of homogeneity of variance in 1 way between groups ANOVA

#1
Hi, I'm a qualitative researcher trying to get to grips with quantitative data so please be patient with my probably very simple questions! :)
The purpose of my study is to identify if there is any difference between undergraduate year of study (yrs 1,2 and 3) and their assessment and feedback engagement.

The data was collected as a questionnaire, identifying year of study (1,2,3) against other questions about student assessment and feedback, which were on a 5 point Likert scale. I had 182 respondents.

I am using SPSS and think that the 1 way between groups ANOVA was the appropriate test - First question, is this the correct test with this data?

Secondly, using this test I have identified a number of significant differences across some questions, however, I noticed on two questions that the Levene's test of homogeneity of variances is violated, but I am not sure what to do about that. Reading around it suggests moving to a non parametric test, but it seems strange to test the same data in different ways (e.g. some questions as the ANOVA and others as a non parametric test due to the Levene's violation? Any advice as to how to deal with this issue would be very much appreciated. Or indeed, if you feel there is a more appropriate test I should be using! :)

Many thanks,
Tig
 

Karabiner

TS Contributor
#2
questions about student assessment and feedback, which were on a 5 point Likert scale.
The term Likert scale denotes a scale which consists of several Likert-type items.
The 5 point response scale of a single item is not a Likert scale.

I am using SPSS and think that the 1 way between groups ANOVA was the appropriate test - First question, is this the correct test with this data?
Do you want to analyse Likert-type items (ordinal scaled), or Likert scales (interval scaled)?
If only items, then oneway ANOVA would be inappropriate. Rather, you should use
Kruskal-Wallis H-test.

But since your grouping variable (study year) is ordinal and not just categorical, a Spearman rank
correlation would be the better option, since it would make better use of the information contained
in the "study year" measurement (assumed that you investigate monotonous relationships like
"the higher the study year, the higher [or: the lower] the score").

With kind regards

Karabiner
 
#3
Thank you for your reply,

The term Likert scale denotes a scale which consists of several Likert-type items.
The 5 point response scale of a single item is not a Likert scale.
I asked three different groups (students in year 1, 2 and 3) to answer to what extend they agree or disagree with about 20 statements. (I used a 5 point likert scale: I completely agree through to completely disagree) - I assumed this is an ordinal scale, is this correct?

But since your grouping variable (study year) is ordinal and not just categorical, a Spearman rank correlation would be the better option
OK, I had the year of study in SPSS as nominal, so I should change that to ordinal?

I have attached my SPSS file if this helps to understand the data more as I realise I am a complete statistical muppet! :)

Thanks again,
Tig
 
Last edited:

Karabiner

TS Contributor
#4
(I used a 5 point likert scale: I completely agree through to completely disagree)
This is the response format of a Likert-type item.
A Likert scale consists of several such items which are summed up.

I assumed this is an ordinal scale, is this correct?
Yes. Therefore it cannot be analysed using analysis of variance (ANOVA),
because ANOVA assumes an intervals scaled dependent variable.

OK, I had the year of study in SPSS as nominal, so I should change that to ordinal?
If you want to analyse monotonous relationships between study year and
the responses to the Likert items, then this would make sense.

With kind regards

Karabiner