In: Statistics and Probability
A cognitive psychologist is testing the theory that short-term memory is mediated by subvocal rehearsal. This theory can be tested by reading aloud a string of letters to a participant, who must repeat the string correctly after a brief delay. If the theory is correct, there will be more errors when the list contains letters that sound alike (e.g., G and T) than when the list contains letters that look alike (e.g., P and R). Each participant gets both types of
letter strings, which are randomly mixed in the same experimental session. The number of errors for each type of letter string for each participant are shown in the following table:
Conduct the appropriate t-test (independent or paired) to test the hypothesis that more errors occur when the list contains letters that sound alike than when the list contains letters that look alike using a one-tailed α = .05.
Participant Letters sound alike Letters look alike
1 |
8 |
4 |
2 |
5 |
5 |
3 |
6 |
3 |
4 |
10 |
11 |
5 |
3 |
2 |
6 |
4 |
6 |
7 |
7 |
4 |
8 |
11 |
6 |
9 |
9 |
7 |
Calculate Cohen’s d as a measure of effect size, and state whether the effect is small, medium, or large.
Calculate the 99% confidence interval to determine whether or not the result is significant at the the α = .01 level.
Report the your results in APA style as you would do in a journal article.
Solution:
Here, we have to use paired t test.
H0: µD = 0 versus Ha: µD > 0 (upper/right/one tailed test)
From given data, we have
Dbar = 1.6667
SD = 2.3452
n = 9
df = n – 1 = 8
Level of significance = 0.05
Critical t value = 1.8595 (by using t-table or excel)
Test statistic = t = (Dbar - µD) /[SD /sqrt(n)]
t = (1.6667 – 0) / [2.3452/sqrt(9)]
t = 1.6667/(2.3452/3)
t = 2.132056967
P-value = 0.0328
P-value < α = 0.05
So, we reject the null hypothesis H0
There is sufficient evidence to conclude that more errors occur when the list contains letters that sound alike than when the list contains letters that look alike.
Cohen’s d = Dbar/SD = 1.6667/2.3452 = 0.710686
Effect is medium.
Effect is small if d≤0.2, medium if d lies between 0.2 to 0.8, large if d≥0.8.
Confidence interval = Dbar ± t*[SD /sqrt(n)]
Confidence level = 99%, df = 8, t = 3.3554 (by using t-table or excel)
Confidence interval = 1.6667 ± 3.3554*[2.3452 /sqrt(9)]
Confidence interval = 1.6667 ± 3.3554* 0.7817
Confidence interval = 1.6667 ± 2.62291618
Lower limit = 1.6667 - 2.62291618 = -0.95621618
Upper limit = 1.6667 + 2.62291618 = 4.28961618
The difference ‘0’ is lies within above confidence interval, so we conclude that result is not significant at α = 0.01 level.