Statistical Power, Limitations Of T-tests + Alternatives - Week 8

Question 1

Define statistical error

Answer 1

The probability that the test correctly rejects H0 when H1 is true
(Effect exists in the population)

Question 2

What three things affect power levels?

Answer 2

Alpha Level
Effect size
Sample Size

(Higher = more power = more significant results)

Question 3

How can we increase Cohens D? (3 ways)

Answer 3

Use a strong manipulation (increase M1-M2)

Reduce within group variability : homogenous sample (decrease SD)

Have standardised procedures and cautious measurements

Question 4

Advantages of homogenous samples

Answer 4

Limit population - reduced within-group variability - increased Cohens D - More statistical power

Question 5

Advantages of heterogenous samples

Answer 5

Capture widest range of population - lowers Cohens D - increases observed correlations

Question 6

Relationship between α and statistical power

Answer 6

Statistical power increases as α increases

Question 7

Why do we use α = 0.05 in Psychology experiments? And how can we overcome limitations?

Answer 7

Less type I errors
Lower levels of power - must increase N to overcome this

Question 8

2 main importances of statistical power

Answer 8

In studies - what N we need for sufficient power

How to interpret non-significant results: if small N and non-sig results = lower power = miss medium sized effects

Question 9

What is the level of statistical power we aim to use in Psychology?

Answer 9

80%

Question 10

Assumptions of Paired T-Tests

Answer 10

Interval / Ratio scale
Populations have normal distributions

Question 11

Assumptions of independent T-Tests

Answer 11

Additional of populations with equal variance

Question 12

What happens when violations of assumptions are present?

Answer 12

[If assumptions met α = probability]

Unequal variance = Type I Errors (t-test can’t be used)

Question 13

When can t-tests not be used?

Answer 13

Presence of strongly skewed data
(Independent only) : BOTH extreme unequal valences and unequal N

Question 14

What can we use as a method of data transformation when data is strongly skewed to allow a t-test?

Answer 14

Log Transformations
- Ranks data to reduce num of data points between
-Standardises to allow for a t-test to be used

Question 15

What are the two types of non-parametric tests and what type of groups are needed for their use?

Answer 15

Mann Whitney U-Test (Independent Groups)
Wilcoxin Test (Paired Groups)

Both use ranking (no skew and outliners) and test H0 that medians of relevant positions are the same

Question 16

How is statistical power calculated?

Answer 16

1 - β
Where β = Type II Errors