Segment 1: Actor-Partner Models vs. Common-Fate Models

Slides:



Advertisements
Similar presentations
Seven Deadly Sins of Dyadic Data Analysis David A. Kenny February 14, 2013.
Advertisements

Structural Equation Modeling
Maternal Psychological Control: Links to Close Friendship and Depression in Early Adolescence Heather L. Tencer Jessica R. Meyer Felicia D. Hall University.
Outline 1) Objectives 2) Model representation 3) Assumptions 4) Data type requirement 5) Steps for solving problem 6) A hypothetical example Path Analysis.
APIM with Distinguishable Dyads: SEM Estimation
“Ghost Chasing”: Demystifying Latent Variables and SEM
Week 14 Chapter 16 – Partial Correlation and Multiple Regression and Correlation.
Module 32: Multiple Regression This module reviews simple linear regression and then discusses multiple regression. The next module contains several examples.
LEARNING PROGRAMME Hypothesis testing Intermediate Training in Quantitative Analysis Bangkok November 2007.
Moderation in Structural Equation Modeling: Specification, Estimation, and Interpretation Using Quadratic Structural Equations Jeffrey R. Edwards University.
CJT 765: Structural Equation Modeling Class 7: fitting a model, fit indices, comparingmodels, statistical power.
Disentangling the Relations between Discrimination, Cultural Orientation, Social Support, and Coping in Mexican American Adolescents Megan O’Donnell Mark.
Illustrating DyadR Using the Truth & Bias Model
SEM: Basics Byrne Chapter 1 Tabachnick SEM
Ethnic Identity among Mexican American Adolescents: The Role of Maternal Cultural Values and Parenting Practices 1 Miriam M. Martinez, 1 Gustavo Carlo,
Maternal Romantic Relationship Quality, Parenting Stress and Child Outcomes: A Mediational Model Christine R. Keeports, Nicole J. Holmberg, & Laura D.
1 Analysis Consequences of Dependent Measurement Problems in Research on Older Couples Jason T. Newsom Institute on Aging Portland State University Presented.
ANOVA and Linear Regression ScWk 242 – Week 13 Slides.
Measurement Models: Exploratory and Confirmatory Factor Analysis James G. Anderson, Ph.D. Purdue University.
CJT 765: Structural Equation Modeling Class 12: Wrap Up: Latent Growth Models, Pitfalls, Critique and Future Directions for SEM.
Aron, Aron, & Coups, Statistics for the Behavioral and Social Sciences: A Brief Course (3e), © 2005 Prentice Hall Chapter 12 Making Sense of Advanced Statistical.
Youth violence exposure, adolescent delinquency and anxiety, and the potential mediating role of sleep problems during middle childhood Chelsea M. Weaver.
SEM Basics 2 Byrne Chapter 2 Kline pg 7-15, 50-51, ,
Janis L. Whitlock Cornell University.   Previous research show that human beings develop in multiple social ecologies but school connectedness and the.
Actor-Partner Interdependence Model or APIM. APIM A model that simultaneously estimates actor and partner effects on an outcome variable The actor and.
Marital Satisfaction and Consensus: Links to the Development of Behavioral Social Functioning in Early Adolescence L. Wrenn Thompson Jessica Meyer Joseph.
Chapter 17 STRUCTURAL EQUATION MODELING. Structural Equation Modeling (SEM)  Relatively new statistical technique used to test theoretical or causal.
Satisfaction, Guaranteed: My Perceptions of You Are More Predictive of Negotiation Satisfaction Than Your Actions Devin E. Howington and Sara D. Hodges.
Reciprocal Relations Between Parent-Child Relationship Quality and Children's Adjustment During Early Childhood Chelsea M. Weaver, Anne M. Gill, Katelyn.
T Relationships do matter: Understanding how nurse-physician relationships can impact patient care outcomes Sandra L. Siedlecki PhD RN CNS.
Main effect of “you” category words, F(2, 333)= 24.52, p
Deep Dyadic Friendships vs. Broad Peer Preference During Adolescence as Predictors of Adolescent and Adult Internalizing Symptoms Rachel K. Narr & Joseph.
Stats Methods at IC Lecture 3: Regression.
Structural Equation Modeling using MPlus
Karin Karako Hunter college, the city university of new york
Christian Hahn, M.Sc. & Lorne Campbell, PhD
Introduction Hypotheses Results Discussion Method
Section 2: Science as a Process
Hypothesis Testing and Confidence Intervals (Part 1): Using the Standard Normal Lecture 8 Justin Kern October 10 and 12, 2017.
Two-wave Two-variable Models
Claire A. Wood1, Heather M. Helms2, & W. Roger Mills-Koonce2
CJT 765: Structural Equation Modeling
Introduction Results Hypotheses Discussion Method
Week 14 Chapter 16 – Partial Correlation and Multiple Regression and Correlation.
Participants and Procedures
Self-discrepancies in the Social Role of Mother: Associations between Self-discrepancies and Negative Affect Nicole J. Holmberg, Laura D. Pittman, Emily.
Making Sense of Advanced Statistical Procedures in Research Articles
APIM with Indistinguishable Dyads: SEM Estimation
Office of Education Improvement and Innovation
To obtain a copy of this poster, please visit
A New Approach to the Study of Teams: The GAPIM
Introduction Discussion Results Method References
Introduction Results Methods Conclusions
Introduction Results Methods Conclusions
Cross Sectional Designs
Hypothesis Construction
University of Virginia1 & James Madison University2
The Importance of Positive Peer Relationships in Predicting Decreases in Adolescents’ Depressive Symptoms over Time Joanna M. Chango, Erin M. Miga, & Joseph.
Prosocial Behaviors in Adolescence
Product moment correlation
Confirmatory Factor Analysis
General Social Competence (18)
Korey F. Beckwith & David E. Szwedo James Madison University
The Effects of Childhood Emotional Abuse on Later Romantic Relationship Outcomes: The Moderating Role of Self-Worth, Alcohol, and Jealousy Madeline M.
Kristin E. Gross & David E. Szwedo James Madison University
Regression Analysis.
Psych 231: Research Methods in Psychology
Aashna A. Dhayagude & David E. Szwedo James Madison University
Rachael Bedford Mplus: Longitudinal Analysis Workshop 23/06/2015
MGS 3100 Business Analysis Regression Feb 18, 2016
Presentation transcript:

Segment 1: Actor-Partner Models vs. Common-Fate Models Considering the Unit of Analysis in Dyadic Methods Erin Kramer Holmes, Adam M. Galovan, and Christine M. Proulx

Gonzalez and Griffin (2012) note that in research about relationships: “Interdependence is not treated as a nuisance that needs to be corrected but rather as one of the key psychological parameters to model” (p. 439). This presentation is grounded in this assumption: that interdependence is something worth of study and we need to find methods that allow us to study it well.

Actor/partner Interdependence Model (APIM) Mother Conflict Mother Marital Quality This is one way of modeling dyadic/family level data. It is probably the most common. It identifies actor effects and partner effects, reflecting potential reciprocity in relationships and reflecting the way individuals might influence their partner’s outcomes when they are in a close relationship. Father Conflict Father Marital Quality

Theory behind Studying Dyadic/Family Phenomena Interpersonal interaction is a fundamental component of human life worth exploring. Human thoughts, emotions, and actions are impacted not only by the individual displaying these outcomes, but also by those connected to the individual in his or her life (e.g. romantic partner, friend, child, etc). Processes within the group can operate in direct and indirect ways. These direct and indirect processes represent unique group-level contexts (e.g. common expectations) which may impact other group or individual level phenomena. Individual functioning is related not only to the individuals themselves, but also to the complex system of behaviors between members of the system (e.g. family rules, family goals, roles within the system, etc.) The first two assumptions can be tested with the APIM. But the third brings up a few more considerations. These processes at the group level may include implicit or explicit rules for all members of the system, roles that are assigned and filled within that unique system, behaviors experienced and perpetuated within the system, or expectations shared by multiple members of the group. As such, researchers still need to account for non-independence in the system, assuming that group-level phenomena specific to that system may impact other group-level phenomena, and/or other individual level phenomena that also occur within that system.

Two Methodological Issues to think about Multiple individuals in the same group may share similar responses about the same group-level phenomenon (i.e. shared variance and non-independence of data) Properties of interdependent relationships may represent either individual-level or group-level phenomena “I am happy with my relationship;” “My partner listens to me.” “It is a real zoo in our home”; “We can usually find things when we need them”; “In our relationship, we talk about things that make us angry.”

Conceptual Representation of Handling of Shared variance in the APIM, CFM, and Hybrid Models Mother Mother Father Father Shared Shared [This slide features several animation sequences.] As conceptually represented here, in the APIM the shared variance in actor and partner scores is modeled as correlated error/residual error [SHOWN IN PURPLE HERE], with paths considering only the unique effect of each partner on the unique variance in self and partner outcomes [SHOWN IN RED AND BLUE HERE]. Conversely, in the CFM the shared variance [IN PURPLE] is of the most interest and the remaining variance in each individual’s score is treated as error in the model. Paths consider how the shared variance for the independent variables is related to the shared variance for the dependent variables. Hybrid Models consider either how the shared variance is related to variance in individual outcomes [first set of Red Arrows after the Purple arrow] OR how variance in individual outcomes is related to variance in shared outcomes [second set of Red Arrows after Purple arrow]. [Red and Blue Arrows disappear and Purple Arrow appears on click.]

USING the Common Fate Model “The common-fate conception implies that two dyad members are similar to one another on a given variable due to the influence of a shared or dyadic latent variable” (Ledermann & Kenny, 2012, p. 141). Measure (or items) with wording representing group-level rather than individual- level process “It is a real zoo in our home”; “We can usually find things when we need them”; “In our relationship, we talk about things that make us angry” Assessment of similarity between members (when r > .2, or factor loadings on common latent variable are > .4 see Ledermann & Kenny, 2012; or Ledermann & Macho 2009) If in an APIM the actor and partner effects have the same sign, and there is positive non-independence in both variable pairs, the CFM should yield reliable estimates. Ex. Because we share the same relationship the common fate model assumes that the two of us will report similar experiences. It models a variable that represents our shared variance, putting our unique variance in the error terms.

What Types of Research Questions are Answered by the CFM models? Many questions we have in family research, can be conceptualized as dyad- or family-level questions How are couples’ shared perceptions of conflict and coparenting quality related to their shared view of marital quality? More simply: How is couple conflict and coparenting quality related to marital quality at the couple-level? How is positive coparenting related to child behavior outcomes? How is family chaos related to couple conflict resolution?

Building your common fate model Mother Conflict Father Conflict Mother Marital Quality Father Marital Quality Couple Conflict Couple Marital Quality 1 Discuss briefly Structural Equation Models for those who aren’t as familiar. Parameters: One intercept for each indicator, one variance for each error term, one variance for the exogenous latent variable and one for the disturbance variance for the endogenous latent variable, two covariances between the error terms, and one direct path at the dyad-level. Remember, both of the paths from your latent variable to your observed variables need to be fixed to unity (i.e., the path weight is set equal to 1; see Macho & Ledermann, 2012). As relations between family- or couple-level constructs may be inflated due to correlations between each partner’s individual reports—an idea referred to as intrapersonal dyadic dependence (see Peugh, DiLillo, & Panuzio, 2013)—disturbance terms for each partner’s report of one family- or couple-level variable were correlated with their own report of other family- or couple-level variables. Peugh, DiLillo, & Panuzio (2013) set only one of the indicator path weights to 1. However, this often results in unidentified models. If you want to estimate a path for one of the indicator variables, you can also estimate the latent mean for the common fate variable and set the intercepts for each indicator to zero. In terms of the structural model, the paths between common fate variables are usually not substantively different. We use Macho & Ledermann’s (2012) method in all of our examples. 1 1 1 Create your latent variables (representing how the observed variables are related to the latent group-level constructs).

Example with Generated Data The data were generated in Mplus by using summary statistics such as means, standard deviations, correlations, reliabilities, and factor loadings either from data sets managed by the authors or from published articles Busby et al (1995) Feinberg, Brown, & Kan (2012) After generating a dataset with the proper covariance structure, data were rescaled to fit a set metric Coparenting items range from 0 to 6 Conflict items range from 1 to 5 The Consensus, Satisfaction, Cohesion, and BMI items were not rescaled Higher coparenting scores denote more effective coparenting. Higher conflict scores denote more conflict. Higher Consensus, Satisfaction, and Cohesion scores indicate more positive Marital Adjustment.

Example with the Generated Data Mother Conflict Father Conflict Mother Mar Adj Father Mar Adj Couple Conflict Couple Marital Adjustment .68 .51 –.15 .71 Mother Coparenting Father Coparenting Coparenting .74 .12 .03 –.10 –.54 .58 –.04 –.09 .13 For this example, we constructed the CFM variables as second-order latent variables, with each indicator to the latent shared variable also being a latent variable. After a click, the structural paths change to show that the effect of Conflict on Marital Adjustment is fully mediated by Couple Coparenting. Bootstrap analyses showed this indirect effect was significant (p < .05). .00 –.56 .68 N = 1000. Model Fit Statistics: χ2 (393) = 408.508, ns; CFI = .999; TLI = .999; RMSEA = .006.

Chaos, conflict Resolution, and Child Behavior Problems SUBSTANTIVE EXAMPLE Chaos, conflict Resolution, and Child Behavior Problems

Substantive Example: sample 732 couples and a target child All participants in the Study of Early Child Care and Youth Development (SECCYD) 83% White, non-Hispanic 7.4% African American 5.2% Hispanic 4.4% “Other” Ethnicity 373 male children, 359 female children Data here represent 3 periods in time: 3rd grade, 5th grade, and age 15

hypotheses We hypothesize that greater chaos in the home will result in a less positive emotional tone following conflict. Less positive resolution of conflict between parents will also be associated with child behavior problems. We further hypothesize that associations between chaos in the home and child behavior problems are likely mediated by the impact of chaos on parental conflict resolution.

Measures Sample CHAOS items: You can’t hear yourself think in our home Chaos in the Home Confusion, Hubbub, and Disorder Scale (CHAOS), (Matheny, Wachs, & Phillips, 1995) 15 items assess routine, noise, and confusion. Higher scores represent a more chaotic home environment Conflict Resolution Conflict Resolution Scale, (Kerig, 1996) 13 items designed to assess the “emotional tone” following conflict Higher scores represent a more positive emotional tone following conflict. Children’s Problem Behavior Child Behavior Check List (Achenbach, 1991) 118 items (internalizing, externalizing, and other thought and behavioral problems) Scores on the CBCL are standardized and reported as T-Scores with higher scores indicating more problem behavior. Sample CHAOS items: You can’t hear yourself think in our home It’s a real “zoo” in our home We are usually able to stay on top of things α = .78 for fathers and α = .81 for mothers Sample Conflict Resolution items: We feel closer to one another than before the fight We don’t resolve the issue We continue to hold grudges We each give in a little bit to the other α = .87 for fathers and α = .88 for mothers. CBCL α = .95 for fathers and α = .94 for mothers

Mediation Model with Three Reports on the CBCL Mother Chaos Father Chaos Mother CBCL Father CBCL Family Chaos CBCL .68 .72 .26 .51 .48 3rd Grade Age 15 Mother Conflict Resolution Father Conflict Resolution Couple Conflict Resolution .70 .75 6th Grade –.21 –.23 –.33 –.28 –.15 –.06 Youth CBCL .61 .38 .15 As is evident with the CBCL variable, the CFM model is not limited to only dyadic data. CFM variables can estimated using reports from 3 or more family members. Once the child is added to the CBCL, mother and father scores can be correlated, as they are likely to share variance beyond what is shared with the child. N = 732. Control variables include Father is Partner, Income-to-Needs ratio, Mother’s Education, Child’s Sex, Child is Firstborn, Child is Ethnic Minority, and Mother’s Age. Model Fit Statistics: χ2 (36) = 66.956, ns; CFI = .980; TLI = .943; RMSEA = .034.

Discussion of Substantive Example It is noteworthy that the reports of chaos in third grade were associated with outcomes in both fifth and grade and age 15. According to family systems theory, once families establish patterns of interaction these patterns are maintained by the system. Any effort to change these patterns would be met with resistance, concepts known as homeostasis and negative feedback (S. Minuchin, 1974). By the time a child is in third grade, family interaction patterns are well- established. Thus, patterns of chaos would be established and likely to persist up through the later grades.

Methodological extensions Dyer, Day and Harper (2014) and Gustavson et al (2012) argue that CFM models allow the researcher to explore both shared and unique perceptions “In addition, these errors almost certainly contain a component of random error” (Dyer et al, 2014, p. 4). Unique Perceptions Thus, when predicting both unique and shared perceptions in a CFM model, an additional error term is required. References Dyer, W. J., Day, R. D., & Harper, J. M. (2014, July 7). Father Involvement: Identifying and Predicting Family Members’ Shared and Unique Perceptions. Journal of Family Psychology. Advance online publication. http://dx.doi.org/10.1037/a0036903 Gustavson, K., Røysamb, E., von Soest, T., Helland, M., Karevold, E., & Mathiesen, K. (2012). Reciprocal longitudinal associations between depressive symptoms and romantic partners’ synchronized view of relationship quality. Journal of Social and Personal Relationships, 29(6), 776-794. doi:10.1177/0265407512448264 Father Score Mother Score Shared Perceptions

Models of Shared and Unique Perceptions To explore predictors of both shared and unique perceptions of father involvement, Dyer et al. used CFM techniques and then saved the factor scores and residuals for use in regression models Saving latent variable scores often results in decreased variance, which may lead to biased parameter estimates for analyses that use these parameters (Hoshino & Bentler, 2013) An alternative to saving factor scores, would be to estimate the unique and shared perceptions within a latent variable framework Gustavson et al (2012) employed this method to explore longitudinal relations among partners synchronized (shared) and unique views of relationship quality and each spouses’ depressive symptoms To explore predictors of both shared and unique perceptions of father involvement, Dyer et al. used CFM techniques and then saved the factor scores and residuals for use in regression models. Hoshino and Bentler (2013) note, however, that saving latent variable scores often results in decreased variance, which may lead to biased parameter estimates for analyses that use these parameters. An alternative to saving factor scores, would be to estimate the unique and shared perceptions within a latent variable framework. Gustavson et al (2012) employed this method to explore longitudinal relations among partners synchronized (shared) and unique views of relationship quality and each spouses’ depressive symptoms. Reference Hoshino, T., & Bentler, P. M. (2013). Bias in Factor Score Regression and a Simple Solution. In A. R. de Leon & K. C. Chough (Eds.) Analysis of mixed data: Methods and application (pp. 43-62). New York: Chapman and Hall.

Challenges in Models of Shared and Unique Perceptions It is often not possible to simultaneously estimate predictors of both individual and shared perceptions, as such models are unidentified. We propose a three- step process. CFM model predicting only the shared perceptions is estimated. Parameter estimates for the paths predicting the CFM outcome are set to the (unstandardized) values obtained in step one, and paths predicting individual perceptions are estimated. To verify that stability of estimates, the paths predicting the individual perceptions are set to the values obtained in step two, and the paths predicting the shared perceptions are freely estimated. The estimates for these freed paths should be identical to those obtained in step one. It is not possible, however, to simultaneously estimate both predictors of both individual and shared perceptions, as such models are unidentified. Gustavson et al (2012) did not indicate how they handled these model identification issues. We propose a three-step process. First, a CFM model predicting the shared perceptions is estimated. Next, parameter estimates for the paths predicting the CFM outcome are set to the values obtained in step one, and paths predicting individual perceptions are estimated. Finally, to verify that stability of estimates, the paths predicting the individual perceptions are set to the values obtained in step two, and the paths predicting the shared perceptions are freely estimated. The estimates for these freed paths should be identical to those obtained in step one. This process essentially follows the procedure outlined by Dyer et al (2014)—as saving the residual scores and using them as dependent variables in later analyses assumes that the original regression paths predicting the shared perceptions is constant—but retains the advantage of latent variable models.

Mediation Model with Shared and Unique Perceptions Mother Chaos Father Chaos Mother CBCL Father CBCL Family Chaos CBCL .69 .73 .28 .47 3rd Grade Age 15 Mother Conflict Resolution Father Conflict Resolution Couple Conflict Resolution .70 .74 6th Grade –.22 –.27 –.33 –.17 –.02 Youth CBCL .64 .40 .16 .13 Youth Unique CBCL Mom Unique CBCL Dad Unique CBCL Unique CBCL predictor paths: Family Chaos  Mom Unique CBCL = .03, ns Family Chaos  Dad Unique CBCL = .00, ns Family Chaos  Youth Unique CBCL = –.05, ns Couple Conflict Resolution  Mom Unique CBCL = –.03, ns Couple Conflict Resolution  Dad Unique CBCL = .01, ns Couple Conflict Resolution  Youth Unique CBCL = .05, ns N = 732. Control variables include Father is Partner, Income-to-Needs ratio, Mother’s Education, Child’s Sex, Child is Firstborn, Child is Ethnic Minority, and Mother’s Age. Model Fit Statistics: χ2 (32) = 65.200, p < .01; CFI = .979; TLI = .931; RMSEA = .038. Family Chaos  Mom Unique CBCL = .03, ns Family Chaos  Dad Unique CBCL = .00, ns Family Chaos  Youth Unique CBCL = –.05, ns Couple Conflict Resolution  Mom Unique CBCL = –.03, ns Couple Conflict Resolution  Dad Unique CBCL = .01, ns Couple Conflict Resolution  Youth Unique CBCL = .05, ns

Limitations Estimation of CFM can sometimes be difficult There can be issues with non-identification Theoretically, items should be at a group- or family-level rather than the individual level For example, it doesn’t make a lot of sense to model husband and wife depression as “couple depression” As a field, we have many names for these types of models Common Fate Model Multi-informant model Multi-level dyadic structural equation model What is the difference between each? OTHER LIMITATIONS???

Implications of CFM Modeling As a family studies field, we often ask questions about processes at the family- or dyad-level. Such questions may often be answered more correctly with analytical techniques that assess dyadic or family variables and processes, such as the CFM approach. One of the challenges in using the CFM may be finding measures that are suited to assessing dyadic and family-level constructs. In some cases, it might be appropriate to modify existing measures to ask questions using plural pronouns (e.g., we, our) rather than singular pronouns (e.g., I, my). These measures may then better assess the dyadic- or family-construct rather than an individual’s perception. In other cases, modifying pronoun usage may not be sufficient. Interdependence or systems frameworks may suggest that constructs scholars have traditionally considered at the individual-level deserve reconceptualization at the dyad- or family-level. In some cases, it may be acceptable to use the CFM even when constructs are not assessed with plural language if the language seems to be clearly directed at assessing a dyadic- or family-level construct and the reporter scores are sufficiently correlated. For example, partners may report on their participation in housework by rating their own contribution in comparison to their partner. Even though partners may be reporting their own contribution, they are reporting on how housework is shared—a dyadic-level construct—and thus each partner’s score could be coded and used as a CFM indicator.

Conclusion As a field focused on studying families and close relationships, we should more often assess constructs at the dyad- and family-level and employ suitable methods for analyzing such data, such as the CFM. The use of these methods will better align with our theoretical assumptions and add greater understanding to the base of research in many substantive areas of family studies.

Questions?

Acknowledgements We are grateful to the Eunice Kennedy Shriver National Institute of Child Health and Human Development Early Child Care Research Network for designing and carrying out the data collection for the example in this project. The NICHD Study of Early Child Care is a study directed by a Steering Committee and supported by NICHD through a cooperative agreement that calls for scientific collaboration between the grantees and the NICHD staff. The content of this project is solely the responsibility of the named authors and does not represent the official views of the Eunice Kennedy Shriver National Institute of Child Health and Human Development, the National Institute of Health, or individual members of the Network.