Trait Approaches and Psychometric Validation
Individual Differences and Assessment
- Behavioral approaches (Skinner, Dollard, and Miller).
- Social cognitive approaches (Bandura, Mischel).
- Assessment methods:
- Observation (self/others).
- Questionnaires.
- Test-retest reliability.
- Item-total reliability.
Psychosocial Problem Conceptualization
- Challenge: Comparing two measures of social anxiety.
- Application: Measurement of key constructs.
- Key psychometric notions:
- Item-total correlations.
- Test-retest reliability.
- Construct validity.
- Diaries.
Cognitive Behavioral (CB) Approaches
- Finishing off CB approaches.
- Trait approaches (Cattell, Eysenck).
- Assessment:
- Questionnaires.
- Item-total reliability.
- Intra-class correlations.
- Factor structure.
- Interpreting fit.
- Criterion validity.
Cognitive Behavioral Theory
- Expand on “hot cognitions” (Mischel’s Cognitive Affective Unit).
- Draw on earlier social cognitive theorists like Bandura and Ellis.
- Core beliefs:
- Drive behavior.
- Are deeply held and resistant to change.
Albert Ellis and Rational Emotive Therapy
- A focused form of social cognitive theory.
- Specific beliefs that drive mood/behavior:
- Demandingness:
- ”I have to be viewed favorably by people that matter to me.”
- ”I must not be dismissed by my peers.”
- Awfulizing:
- ”It's awful if others do not approve of me.”
- ”It's terrible if my team doesn’t respect me.”
- Low-frustration tolerance:
- ”I can’t tolerate failing.”
- ”I can’t bear not getting better at what I do.”
- Depreciation:
- ”If others think I am no good at what I do, I am worthless.”
- ”I am a loser if I do not succeed at things that matter to me.”
Common Core Beliefs About Self
- Unlovability: Something wrong with me.
- Helplessness:
- Being ineffective in getting things done (I am incompetent).
- Vulnerability (I am likely to get hurt).
- Ineffective compared to other people (I am inferior, I can’t achieve like others).
- Worthlessness:
- I am a bad immoral sinner.
- I am toxic bad person.
- Negative beliefs about the world:
- Other people are dangerous, don’t care.
Construct Validity
- At least 5 types of construct validity:
- Concurrent validity: New measure correlates well with previously established similar measure, administered at the same time.
- Convergent validity: New measure correlates well with another measure of a theoretically related construct, administered at the same time.
- Criterion-related validity: Predicting group membership or a real-world category.
- Predictive validity: Does the measure predict something in the future that you would expect it to predict?
- Divergent validity: A new measure correlates poorly with unrelated constructs.
Standard Deviation (SD)
- A measure of dispersion – the average distance from the mean.
- Is it good to have a SD that is much bigger than the mean?
- Is it good to have a really small SD?
Trait Approaches
- How relevant are they for understanding individual differences?
- Not a great fit (in my view).
- Keep pushing forward with psychometric analyses of measures, using trait theory and measurement tools as an example.
- How to evaluate a measurement tool – factor structure.
Trait Theories Overview
- Person-centered versus situationism.
- They tend not to do well at accounting for development.
- A focus on the present.
- Consistent across time.
- Consistent across context.
- Low emphasis on mechanisms.
- Not concerned with “why” or “how” personality changes.
- Primarily concerned with description, organisation and prediction
Raymond Cattell
- Personality is that which permits a prediction of what a person will do in a given situation.
- Desirable/undesirable, normal/abnormal is irrelevant.
- Rigorous scientific approach: SCIENCE DEMANDS MEASUREMENT!
- Distinguishing feature: FACTOR ANALYSIS.
- 16PF.
Factor Analysis
- A data grouping and data reduction technique.
- Based on the logic of the correlation coefficient.
- Trait theory a good subject to explore with factor analysis.
- A nice paper – easy to read if you are new to it. Williams, Onsman & Brown 2010 (QUT Readings Week 4)
- Aims:
- Reduce number of vars – principle of parsimony.
- Examine the structure of your questionnaire.
- Detect whether unidimensional or multidimensional.
- Evaluate construct validity.
- Can look for multicollinearity.
- Can use to develop/refine theoretical constructs.
- Sample size – hugely variable 1:20 or higher is typical.
- What type of factor extraction?
- Principal components analysis (PCA).
- Principal axis factoring.
- Maximum likelihood.
How to Decide on Number of Factors
- A few methods.
- In CFAs you can propose a certain number – force the model to fit 1 or more factors.
- In EFAs it is often less clear.
- In general, don’t rely on one method.
- Eigenvalues great than 1.
- The Scree Test.
- Can look at how much variance each solution adds.
- Eigenvalues - this value tells you how much variance is explained by a group of items. Each eigenvalue will be greater than 1, but usually factors close to one don’t account for much variance (this approach is overinclusive).
- Extraction loadings – each factor starting with the biggest eigenvalue will add a certain amount of variance (‘cumulative %’) – gets smaller and smaller as you go along… usually stop increasing factors when cumulative % gets really small.
- A plot of the factor/principal component ordered by eigenvalue size – look for change in direction of lines of best fit.
- Variance - a measure of dispersion – the average of squared deviations around a mean.
Hans Eysenck
- Eysenck Personality Inventory.
- Concise model: 3 dimensions, or superfactors…
- Thought that Cattell’s 16 personality factors were unreliable.
- Personality largely governed by biology and shaped in childhood.
The Superfactors
- Extraversion versus introversion (E).
- Neuroticism versus emotional stability (N).
- Psychoticism versus impulse control (P).
- Oriented toward outside world.
- Prefer company.
- Highly sociable, impulsive, assertive, dominant and adventurous.
Neuroticism
- Anxious, depressed, tense, irrational, moody.
- Largely inherited.
- More activity in sympathetic branch of the ANS (bodies alarm system).
- Hypersensitivity and emotionality.
Psychoticism
- Aggressive, antisocial, tough-minded, cold egocentric versus impulse control (warmth and empathy).
- Can be cruel, hostile, insensitive.
- Large genetic component.
Traits/Facets (Costa & McCrae)
- Developed the NEO Personality Inventory – Revised, which included some elements of Eysenck’s model but included two more..
- Each dimension based on an extreme (e.g., extraversion – introversion), each person falling somewhere on the continuum.
Gignac
- Gignac – ‘partial confirmatory factor analysis’.
- Something in between an EFA and a CFA.
- Used when know the number of factors, but don’t know what the loadings are…
Factor Analysis Continued
- Identification of “factors”.
- Factor - cluster of related behaviour measures.
- Factor loading - extent to which each measure is related to each factor.
- Items with high loadings go together.
- Ideally factors should be minimally correlated.
Factor Naming
- Factors are named based on semantic consistency.
- Identification of factors is empirical, naming of factors is rather subjective
Exploratory versus Confirmatory Factor Analysis (EFA vs CFA)
EFA
- An unrestricted FA.
- All latent constructs are correlated with all items.
- Focus is on establishing an underlying factor structure.
- Used for scale development, on items that haven’t been tested much
CFA
- Used to verify a factor structure.
- Based on theoretical reasons and/or past empirical research
- Hypothesis a structure and test how the data fit with that structure.
Second Order FA
- One could do a factor analysis that has two levels….
- Neuroticism: Anxiety, Hostility, Depression, Self- consciousness, Impulsiveness, Vulnerability.
- Extroversion: Warmth, Gregariousness, Assertiveness, Activity, Excitement-Seeking, Positive Emotions.
- Openness to Experience: Fantasy, Aesthetics, Feelings, Actions, Ideas, Values.
- Agreeableness: Trust, Straightforwardness, Altruism, Compliance, Modesty, Tender- mindedness.
- Conscientiousness: Awareness of actions and consequences: Competence, Order, Dutifulness, Achievement Striving, Self-Discipline, Deliberation.
How Well Does a Model Fit?
- Lots of indices to test how well the proposed model accounts for the correlation of all items.
- Chi-square – want small numbers, generally less than 3.
- Fit of the model – don’t want a significant difference (>.05).
- Comparative Fit index (CFI).
- Analysis the discrepancy between the data and hypothesised model, while adjusting for sample size.
- CFI > .90 acceptable.
- Standardised root mean square residual (SRMR).
- Fit of observed correlation matrix and the predicted correlation – positive biased when sample sizes are small.
- want small as possible, generally less than 0.09.
RMSEA
- Root mean square error of approximation (RMSEA).
- A sophisticated fit index tests how far a hypothesised model is from a perfectly fitting model.
- Smaller RMSEA, the better (ideally less than 0.08 for ‘good fit’)’.
- Assessing fit.
- Do not rely on a single measure.
- Look at the pattern of fit across several indices (it is not necessarily bad to have one of several indices that suggests the model doesn’t fit).
Etzler & Rohrmann (2017)
- Development and preliminary validation of a brief questionnaire of psychopathic personality traits’.
- Measuring psychopathology via self-report.
- Psychopathic Personality Inventory (PPI) is well validated, but has 154 items and inconsistent factor analysis, 45-50 minutes.
- This study: Develops the Questionnaire of Psychopathic Personality Traits.
Key Aims of the Paper
- Evaluated whether:
- the measure correlates with other established measures (focus on how correlates with another measure called the PPI) – Kind of construct validity?
- The scale correlates with an external ‘gold standard’ criterion, such as criminal charges? - Kind of construct validity?
- Does the scale have factorial invariance (i.e., do the psychometric properties stay the same across different groups?) - Kind of construct validity??? (more debate here – what do you think?)
Statistical Analyses
- Missing data for the FPP was 2.5%.
- Following Hu and Bentler (1999), model fit was considered acceptable if x2/df<2, RMSEA<0.06, SRMR<0.08, and CFI>0.95. Factor loadings of 0.300 were considered as acceptable.
FPP Constructs
- Lack of empathy, fearlessness, narcissistic egocentrism, impulsivity, social manipulation, power, fearlessness.
- Second order trait ‘Psychopathology’.
Week 4 Summary
- Factor analysis (EFA vs CFA).
- What a FA can tell you about your tool:
- Does it measure what you want it to measure?
- How clear and distinct are the factors?
- How to read a table of factors.
- How to decide on the number of factors.
- How to interpret fit.