Likert and t-test

LSM

New Member
#1
Newbie here...

I administered 2 surveys, pre and post event using a 1-5 Likert-type scale. I am using the t-Test for correlated measures to determine if the results are due to chance.

Since the scale is only 1-5 (very rarely-very often), the differences in my questions to 7 participants are averaging .5 to 1.5 points. Should I be using these points in my calculations? Or, should I be figuring the percentage differences between the 1-5 intervals? Using these points, my t-stat results are very low, i.e.:

t-Test: Paired Two Sample for Means

Variable 1 Variable 2
Mean 2.714285714 3.857142857
Variance 0.904761905 1.476190476
Observations 7 7
Pearson Correlation -0.18541927
Hypothesized Mean Difference 0
df 6
t Stat -1.803950467
P(T<=t) one-tail 0.060643511
t Critical one-tail 1.943180274
P(T<=t) two-tail 0.121287022
t Critical two-tail 2.446911846

Speaking of percentages... on a 1-5 scale would the difference between 2 and 3 be 20% or 25%?
 
#2
Hi LSM,

I would the raw values (points) for the analysis. The output shows that the p-value is greater than .05 for both the one-tailed and two-tailed test. There's no significant difference between pre and post, and the results are due to chance.
 

JohnM

TS Contributor
#4
Likert data can be treated as parametric without much concern over violating assumptions of normality.

...and if you try a nonparametric test such as Mann-Whitney or Wilcoxon on a 5-point scale, you'll almost never get a significant result.