CHP. 5 - Construct Validity Flashcards
3 Common Methods of Measurement
- Self-Report (individual reporting their own behavior)
- Observational (scientists record physical traces of behavior)
- Physiological (record biological data)
Construct Validity
Helps us determine if method used to operationalize variable is reliable & accurate
Quantitative Variables
Coded with meaningful numbers
- ordinal, interval, & ratio scale
- ex. height, weight, scale of well-being
Categorical Variables
One variable is NOT quantitatively “higher” than another
- nominal scale
- levels are distinct categories
- ex. race, specifies, country
Ordinal Scale
Ranked order
- degree of variation between units is NOT equal
Interval Scale
Numerals represent equal distances between levels
- no “true 0”
- ex. IQ, SAT scores, temp in F or C
Ratio Scale
Numerals represent equal intervals & there is a true 0
- 0 = absence of what’s being measured (typically can’t have negative values)
- ex. height, weight, age, temp in K
Item vs. Scale
- Scale refers to the summation of data for all items
- Item refers to an individual question
Likert Item
Question response scored from 1 to 5 (strongly agree to strongly disagree)
Semantic Differential Scale
Measure subjective perception of 2 opposite things
- ex. messy to clean, cold to warm
What must construct validity have?
Reliability & validity
3 Types of Reliability
- Test-retest
- Inter-rater
- Internal
Test-Retest Reliability
Reliability of a measure over time
- relatively stable over time
Inter-Rater Reliability
Scores are the same no matter who measures or observes
Internal Reliability
With-in study consistency
- ex. participants respond with same pattern regardless of how question is rephrased
Correlation Coefficient (r)
Determines effect size (between -1 & 1)
- when r is close to 0 = weak/no effect size
- when r is positive = positive association
- when r is negative = negative association
- closer to extremes = stronger relationship
Kappa (k)
Used when observers are recording categorical variables
- k close to 1 = observers agreed
Cronbach’s Alpha (C-alpha)
Used for internal validity
- C-Alpha closer to 1 = more reliable
Types of Validity
Subjective:
- face
- content
Empirical:
- Criterion
- Convergent
- Discriminate
Face Validity
Extent to which a measure appears to measure the intended construct
- judgement call
- weakest defense for overall construct validity
Content Validity
Does the measure fully capture all parts of a construct?
- capture the “whole pie”
Criterion Validity
Does a new measure correlate with a relevant behavioral outcome?
- can use known groups
- ex. spatial working memory with mice w/hippocampal lesion
Convergent Validity
Measures of the same construct should have strong correlation
- how closely a new scale is related to other variables & other measures of same construct
- ex. spatial working memory with MWM & Barnes maze
Discriminate Validity
Not only should the construct correlate with related variables, but should NOT correlate with dissimilar, unrelated ones
- measures of different constructs should have weak/no correlation
- ex. spatial working memory & attention
