27
Statistics II: Measurement & Data Analysis Brian J. Piper

Research Methods: Measurement

Embed Size (px)

DESCRIPTION

lecture 6 from a college level research methods in psychology course taught in the spring 2012 semester by Brian J. Piper, Ph.D. ([email protected]) at Linfield College, includes categorical, ordinal, interval, and ratio levels

Citation preview

Page 1: Research Methods: Measurement

Statistics II: Measurement & Data Analysis

Brian J. Piper

Page 2: Research Methods: Measurement

Goals

• Operationalization• Psychometrics (Reliability & Validity)• Scales of Measurement• Hypothesis Testing• Advanced Topics– Power– Effect Size

Page 3: Research Methods: Measurement

Operationalization

• What areas of the brain are important for music appreciation?

• Important = Active– EEG (electrical activity)– PET (sugar use)– fMRI (oxygen use)

Page 4: Research Methods: Measurement

Operationalization 1: Spatial Function

• Do men and women differ in their spatial abilities?

• Spatial Function = mental rotation

Page 5: Research Methods: Measurement

Operationalization 2: Motor Function

• Do men and women differ in their fine-motor abilities?

• Motor Function = rotary pursuit

Page 6: Research Methods: Measurement

Reliability

• Consistency of measurement• Types– Test-re-test reliability

Time 1

Time 2

Page 7: Research Methods: Measurement

Split-Half Reliability

• Consistency of measurement on two-halves of test

• Foundations for repeated measurements

Even

Odd

Page 8: Research Methods: Measurement

Extension: Short-Form of Wisconsin (Berg) Card Sorting Test (BCST)

Page 9: Research Methods: Measurement

Extension: Short-Form of Wisconsin (Berg) Card Sorting Test (BCST)

r(205) = +0.77

Fox et al. (in review). J Biol Biomed Reports.

Page 10: Research Methods: Measurement

Validity

• Does a test measure what it claims to?• face “faith” validity: does it seem valid based on

intuition (non-numerical)

Page 11: Research Methods: Measurement

Criterion Validity

• Does performance on new measure match with older “gold standard” measure?

• Continuous Performance Tests Example

Reaction Time (Conner’s)

Reac

tion

Tim

e (P

EBL)

Page 12: Research Methods: Measurement

Construct Validity

• Does a test measure the construct it claims to?

• Convergent Validity: Does test A correlate (converge) with test B?

• Discriminant Validity: Does test A measure something different (discriminate) than test C?

Page 13: Research Methods: Measurement

Measurement Scales (Self-Test)Level Definition

Nominal

Ordinal

Interval

Ratio

Page 14: Research Methods: Measurement

Measurement Scales (Self-Test)Level Definition

Nominal categorical, e.g. sex

Ordinal ranking, e.g. Olympic medal

Interval equal spacing, e.g. IQ, ACT, SAT

Ratio true zero, e.g. Reaction Time

Page 15: Research Methods: Measurement

Hypothesis Testing

• Null hypothesis (H0): A = B

• Alternative hypothesis (HA): A ≠ B

Page 16: Research Methods: Measurement

Alpha

• The cut-off used to decide between H0 and HA

• Probability that finding is not due to chance (p value)

• .05: conventional• .10: liberal (some medical environments)• .01: conservative, large N

Page 17: Research Methods: Measurement

Alpha

P value obtained

Decision

.50 H0

.11 H0

.06 H0

.0500000001 H0

.0499999999 HA

Page 18: Research Methods: Measurement

Decision Making

HO is True HO is False

Fail to reject H0 Correct decision

Reject H0 Correct decision

Reality

Decision

Page 19: Research Methods: Measurement

Decision Making

HO is True HO is False

Fail to reject H0 Correct decision

Reject H0 Type I error Correct decision

Reality

Decision

Type I Error: rejecting H0 when it is true

Page 20: Research Methods: Measurement

Decision Making

HO is True HO is False

Fail to reject H0 Correct decision Type II error

Reject H0 Type I error Correct decision

Reality

Decision

Type I Error: rejecting H0 when it is trueType II Error: fail to reject H0 incorrectly

Page 21: Research Methods: Measurement

Publication Bias

• H0 results often don’t get shared• Reasons: – Journal prestige– Research ego– Higher standard

• Solution: registry?– Replication?

Page 22: Research Methods: Measurement

Solution 1: Effect Size Distribution

• A quantitative index of the magnitude of group difference’s

• Calculated as (Mean1 – Mean2)/SD

# St

udie

s

Page 23: Research Methods: Measurement

Solution 1: Effect Size Distribution

• A quantitative index of the magnitude of group difference’s

• Calculated as (Mean1 – Mean2)/SD

# St

udie

s

# St

udie

s

Page 24: Research Methods: Measurement

Solution 2: Power Analysis

• Power: the probability that a real effect will be detected

• Probability of Type II error: Beta• Power = 1 - Beta

N Power

50 0.40

100 0.70

500 0.80

1000 0.85

Page 25: Research Methods: Measurement

Other Terminology

• Population: all members of identifiable group• Sample: a subset of the population• Confidence Interval: inferential statistic,

contains range of where population mean sits

Page 26: Research Methods: Measurement

Margin of Error

• Is accurate if sample is representative of population.

Page 27: Research Methods: Measurement

Summary• Operationalization• Reliability & Validity (face, criterion, construct)• Scales of Measurement (nominal, ordinal,

interval, ratio)• Hypothesis Testing: Type I versus Type II error• Advanced Topics– Power– Effect Size