1.56k likes | 1.58k Views
This paper discusses new advances in measurement related to relationship quality and responsiveness to change. Topics covered include IRT optimization, bi-dimensional view, implicit measures, screening for error variance, and attention/effort. The paper also acknowledges the work of various researchers and introduces the Couples Satisfaction Index (CSI). The paper concludes with a discussion on the longitudinal analysis and responsiveness to change over time.
E N D
New Advances in Measurement Ronald D. Rogge
TOPICS • RELATIONSHIP QUALITY • T1: IRT Optimization • Study 1 • T2: Responsiveness to Change • Studies 2-5 • T3: Bi-Dimensional View • Studies 6-7 • T4: Implicit Measures • Studies 8-11 • ATTENTION • T5: Screening for Error Variance • Studies 12-16
Acknowledgements • Couples Satisfaction Index • Janette Funk, Mike Maniaci, Maria Saavedra, Soonhee Lee • Positive-Negative Relationship Quality • Frank Fincham, Richard Mattson, Matt Johnson • C.A.R.E. Program • Tom Bradbury, Rebecca Cobb, Matt Johnson, Erika Lawrence, Lisa Story, Lexi Rothman • Implicit Assessment • Soonhee Lee, Harry Reis • Attention / Effort • Mike Maniaci, Janette Funk, Soonhee Lee, Maria Saavedra
Relationship Quality • Relationship satisfaction • Self-report scales (DAS, MAT, QMI) • 30-50 years of research (over 4K studies) • Excellent correlational validity • Level of noise? • Responsive to change over time? • Are these the “best” items?
TOPIC 1: IRT Optimization • Large sample method • N at least 1,000 in smallest group • Large item pool • Unidimensional • Non-redundant • Used by ETS • SAT, GRE, MCAT, LSAT • Quality of each item • Information • Noise • Advantages • Over correlations • Over small sample methods
IRT Approach • Latent scores (q) for each subject • Like GRE scores • Assessing relationship satisfaction • Parameters for each item • Response curves • Higher q’s higher responses? • Item Responsiveness • How informative? • Where informative? • Creates information profiles • For individual items • For sets of items
Study 1 - Measures • 141 satisfaction items: • DAS, MAT, RAS, KMS, QMI, SMD • 71 additional items • 7 anchor scales: • Neuroticism (EPQ-N) • Conflict / Communication (MCI, CPQ, IAI) • Stress (PSS) • Sexual Chemistry (Eros) • Instability (MSI) • 2 validity scales: • Inconsistency (PAI) • Infrequency (PAI)
Study 1 - Sample • 5,315 online respondents • After removing: • Incomplete or invalid responses • Multivariate outliers • 26yo (10yrs) • 83% Female • 76% Caucasian • 26% High school ed. or lower • $27K average income • 24% married, 16% engaged, 60% dating
Evaluating Previous Scales • IRT results • Simultaneous analysis • 66 items of existing scales • Some very informative items • Many poor items
DAS-31(Degree of happiness, all things considered, of your relationship) Response Curves Information Curve
DAS/MAT 5Agreement on: FRIENDS Response Curves Information Curve
MAT 12In leisure time, do you (and does your mate) prefer to be “on the go” or to stay at home? Response Curves Information Curve
From Items to Scales • A scale’s information = sum of information from each item • How informative Across different levels of happiness
Analysis of Existing Measures • Many uninformative items • Particularly for DAS and MAT noise / error • Modest test information • For all scales • Notably poor for MAT and DAS • Room for improvement
Creating the CSI • 141 item pool • Screen for contaminating items • Screen for redundant items • IRT on remaining 66 items • Select 32 most effective
What have we gained? • Identical correlational results • Strong convergent validity • Strong discriminant validity • Strong construct validity Measuring same thing • Higher information… • Should have • Lower Noise • Higher Precision • Greater Power
Satisfaction Groups • IRT satisfaction estimates • For each subject • Based on MAT, DAS, & CSI items • (equivalent of GRE scores) • Created satisfaction groups • N = 265 • HIGHLY similar SAT within each group • MAT, DAS & CSI scores also similar?
Effect Size • Ability to detect difference • Between groups • Pre – Post • Effect Size = M1 – M2. pooled SD • Difference in SD units • Power for detecting D’s in SAT groups
STUDY 1 - Conclusions • CROSS-SECTIONALLY • CSI assess same construct • Higher precision • Higher power • NEXT STEP • Longitudinal analysis • Responsiveness to change over time
TOPIC 2: Responsiveness • Detecting change • Assumption • Longitudinal • External Criteria • Treatment effect • Clinician • Interviewer • Global report • SERM (Sdiff) • Noise over time • Estimating • Two main applications • Individual change • Clinically distinct groups
Studies 2 through 4 • Study 2 • 267 online respondents • 1 & 2wk follow ups • 468 change scores • Study 3 • 156 online respondents • 1, 2, 3, 4, 6 & 12mo follow ups • 455 change scores • Study 4 • 545 online respondents • 1, 2, 3, 4, 6, 9 & 12mo follow ups • 1,552 change scores
Studies 2-4: Measures • Relationship satisfaction scales: • DAS-32 • MAT-15 • CSI-32 • CSI-16* • CSI-4* • 3 global relationship change items • Change since last assessment
Individual Change • How many points of change needed? (to show significant change) • SERM in “No Change” • RCI (Jacobson & Truax, 1991) • MDC95 (Stratford et al., 1996) • MDC95 (SD units) = 1.96*SERM. SD • PRESENTING • Meta-Analytic Summary • Standardized Units
Reliable Individual Change C C C B A
Detecting Change • Individual Change • IRT optimization • Longer scales • Distinct Groups • Can scales distinguish? • Mild deterioration • No change • Mild improvement
Perceived Change • How much have these changed? • Overall happiness in the relationship • Feeling close and connected • Stability of the relationship
Perceived Change • Averaged responses • Alpha = .92 • Created change groups
Quantifying Group-Level Responsiveness • MCID (Guyatt, Walter & Norman, 1987) • Noise over time (SERM) • Effect Sizes: (Avg Change)IMPROVE – (Avg Change)NO CHANGE SERM (Avg Change)DETERIORATE – (Avg Change)NO CHANGE SERM
Analytic Strategy • Improving method • Multi-wave data • Global change continuous • Moderation • HLM • PV: Global change score • Moderators: • Gender • T0 Satisfaction • DV: Change scores on scales (n = 2475) Change scores ≈ 1pt global change MCID effect sizes • Meta-Analytic Summary
Responsiveness in Dissatisfied (1SD below M) A B C C C D C B A A
Responsiveness in Satisfied (1SD above M) A A B B B E D C B A
Responsiveness Conclusions • Can be quantified • Scale selection • Power estimates • Responsive scales • Greater power • Individual • Group • Cross-sectional Longitudinal • Precision & Power translate • NEXT STEP Treatment Effects
Topic 3: Bi-Dimensional View • Uni-Dimensional view • Positive feelings opposite negative feelings • Bi-Dimensional view • Pos/Neg independent • Moderately “dissatisfied” • Ambivalent • Indifferent • Uni-Dimensional obscuring?
Background • Fincham & Linfield (1997) • PN-QIMS • Two 3-item scales • Qualities of spouse • Feelings toward spouse • Feelings about marriage • Considering only (pos/neg) • Separated in survey • CFA in 123 couples • Unique information
Study 5 • Mattson et al. (under review) • New pos-neg scale • 7 SMD items of CSI • Pos / neg separately • Large online sample • Ambivalent • Indifferent
Study 5 - Sample • 1656 online respondents • Demographics • 28yo (7yrs) • 94% Female • 87% Caucasian • 30k income • 5% ≤ high school • Romantic relationships • 38% married (6.5yrs) • 19% engaged (3.6yrs) • 41% dating – exclusive (2.4yrs)
Ambivalence vs. Indifference • Median Splits
Ambivalence vs. Indifference • Median Splits
Study 6 • IRT Optimized Positive & Negative Scales • Item Pools • 20 positive items • 20 negative items • Large sample • UG respondents • Analyses • EFA • Redundancy • IRT • Precision / Power / Validity
Study 6 - Sample • 1,814 undergrad respondents • Demographics • 19yo (2yrs) • 77% Female • 72% Caucasian • Together 2.6yrs • 26% dissatisfied • Close relationships • 54% romantic partners • 38% friends • 5% family members • 3% roommates • Romantic relationships • 76% dating – exclusive • 21% dating – non-exclusive