# SPSS Analysis Question

#### alie104

##### New Member
I'm analyzing data collected from an online cross-sectional survey, using SPSS. There's about 20 questions (and about 150 respondents) on barriers to education. Each question had a respondent rate it on a 0-5 point Likert scale. To answer my research question, "What are the difficulties in accessing education for this population," I'd like to be able to see whether one item was more often rated higher (i.e., closer to 5) in order to see whether it is more often perceived present than the other items. So far, all I can come up with would be desriptive stats analysis, looking at the mean score for each item and seeing which has the highest mean and the highest percentage of "4" or "5" scores. However, a friend suggested an ANOVA analysis to see if there is a significant difference between any of the items. I didn't think that was possible because none of the items have a normal distribution, and technically they're not independent since the same respondents answered for each and every item. Am I missing something, or can anyone think of a different analysis I could do to answer my question? Apologies if this is naive or I'm missing something obvious, I'm a student and relatively new to this.

#### Karabiner

##### TS Contributor
Each question had a respondent rate it on a 0-5 point Likert scale.
Likert scale is not the name of the answering format of Likert-type items, I'm afraid.
Instead, Likert scale is the name of a measurement instrument which consists of
several Likert-type items, which are summed up.

So far, all I can come up with would be desriptive stats analysis, looking at the mean score for each item
Since a 5-point response format is ordinal, the appropriate measure for the central tendency
would be the median. But maybe your instructors will nevertheless accept the mean.

and seeing which has the highest mean and the highest percentage of "4" or "5" scores. However, a friend suggested an ANOVA analysis to see if there is a significant difference between any of the items. I didn't think that was possible because none of the items have a normal distribution, and technically they're not independent since the same respondents answered for each and every item.
Normality is impossible, since you have ordinal scaled items with 6 discrete values.
But even if they were interval scaled, normality of the variables would not matter.

So you want to compare ordinal scaled items which have all been anwered by the
same subjects (i.e. dependent measures, as you correctely noticed). The sign test
would be appropriate for the pairwise comparisons (it'll be 190 pairwise
comparisons).

With kind regards

Karabiner

P.S.
The dependent samples t-test would be used if the variables were interval scaled.
Apart from the question whether the t-test could be used here, please keep in mind
that the dependent samples t-test requires normally distributed differences between
variables (not normally distributed variables). Moreover, if sample size is large enough
(some say n > 30, some n > 50), the t-test is valid even if the differences are not normally
distributed in the population.

#### alie104

##### New Member
Thank you very much Karabiner! I will try the sign test for the pairwise comparisons, and report the median rather than the mean.