Parametric and non parametric tests

Question 1

The type of NHST will depend on…

Answer 1

Number of variables: levels of IVs and DV
Nature of variables: continuous or categorical
Hypotheses being investigated
Type of experimental design
One tailed or two tailed hypothesis

Question 2

Parametric test

Answer 2

Makes assumptions about the population distribution:
The sample IS normally distributed there is no significant skew
Homogeneity of variances: we don’t want major imbalances in sample as a result of participant characteristics so THERE IS EQUAL VARIANCE ACROSS GROUPS
If the DV is interval/ratio level of data
If observations are independent of each other, not influenced by other values in data

Question 3

Non-parametric tests

Answer 3

Do not make assumptions about the population distribution
The sample is not normally distributed Such as if there is significant skew in the data (with a small sample)
if does not reach criteria for parametric:
No homogeneity of variances
Not interval/ratio level data
No independent observations

Question 4

How to check normality of curve

Answer 4

Visually: histograms
Statistically: using z values

Question 5

Checking homogeneity of variance across groups

Answer 5

Assumes roughly equal variaance across groups/ experimental conditions
RUN A LEVENE’S TEST

Question 6

Levene’s test

Answer 6

Checking variances of two samples are approximately equal
When p> .05 the homogeneity of variances can be assumed

Question 7

Welch’s t test

Answer 7

If p< .05 then we violate the variances of samples being approximately equal and they are not due to chance
So we use this test

Question 8

List of null hypothesis test

Answer 8

Independent samples t-test
Paired t test
Wilcoxon signed ranks test
Mann Whitney U test
Chi squared test
Spearman’s rho test
Pearson’s r test

Question 9

Independent samples t-test

Answer 9

One IV at 2 levels
Independent groups (between subjects)
Ratio data

Question 10

Do we accept the null hypothesis?

Answer 10

No, we say we ‘failed to reject’ instead
** if we do reject we say reject

Question 11

if p= .051 when a=.05

Answer 11

We say non significant

Question 12

Do we say insignificant?

Answer 12

No
we ‘ fail to reject the null hypothesis’

Question 13

Writing up results

Answer 13

State in the results section
Unpack in the discussion section

Question 14

Errors

Answer 14

Type 1
Type 2

Question 15

Type 1 error

Answer 15

When we reject the null hypothesis when we shouldn’t have
Because the truth (unknown) is that the null hypothesis is true

Question 16

Type 1 error symbol

Answer 16

a = probability of making a type 1 error
alpha

Question 17

Type 2 error

Answer 17

When we fail to reject the null hypothesis (accept it but dont say that) when we shouldn’t have
Because the truth is that the null hypothesis is false

Question 18

Correct decisions of null hypothesis

Answer 18

The null hypothesis is true in reality:
And we fail to reject it (accept it)
So 1-a (error made when we falsely reject it)

The null hypothesis is false in reality:
And we reject it
So 1-b (error made when we falsely fail to reject it)

Question 19

Type 2 error symbol

Answer 19

B beta
Probability of making a type 2 error

Question 20

Alpha

Answer 20

The odds of saying there is a relationship when there is not one
Aka false positive

Question 21

Beta

Answer 21

Odds of saying no relationship if we assume there is one
Aka false negative/ rejection

Question 22

When choosing statistical tests in terms of error types

Answer 22

Balance possible errors based on what we are measuring eg cancer screening? false positives stay on safe side

Question 23

Reduce risk of errors

Answer 23

Define parameters before study e.g. set alpha and betas= long-run error control
Limit number of hypothesis and statistical significance test (familywise error)
Use bonferroni correction
Replicate results across different studies

Question 24

Familywise error

Answer 24

More statistical tests done, more likely for an error

Question 25

Bonferroni correction

Answer 25

take alpha level and divide by number of tests done