Presentation on theme: "Summer Internship Program Annual Symposium 2012. Agenda Welcome Background Overall Purpose of Symposium Symposium Format Closing Remarks Meet and Greet."— Presentation transcript:
Agenda Welcome Background Overall Purpose of Symposium Symposium Format Closing Remarks Meet and Greet the Interns UM Football Stadium Tour 2
Acknowledgements Sponsors: Health and Retirement Study Life Course Development Program (2) Survey Methodology Program Social Environment and Health Program (2) Partners: Senior Staff Advisory Committee SRC Administrators & SRC Diversity Committee Summer Institute in Survey Research Techniques Survey Research Operations Inter-university Consortium for Political and Social Research ISR and SRC Human Resources SRC Computing ISR and SRC Directors Offices ISR Directors Diversity Advisory Committee 3
The Effects of Incarceration and Probation on Reoffending and Employment Nicole Yadon Social Environment & Health Sponsor: Dr. Jeffrey Morenoff 4
Issues in Studying the Effects of Mass Incarceration Framing the research question and identifying comparison groups Some studies use survey samples to compare people who have vs. have not been to prison We frame the question as being about alternative ways of sanctioning convicted felons Our comparison groups are restricted to the population of people convicted of felonies We compare people who were sentenced to prison, jail, probation, etc. Obtaining appropriate data Survey samples usually dont include institutionalized populations Establishing causality True experiments are not possible – judges will not randomly allocate sentences Problem of unobserved confounders Judges may base their decisions on factors that are not observed by researchers (e.g., temperament) These same factors may predict future outcomes (e.g., recidivism, employment) 7
Our Study Question: What is the effect of being sentenced to prison vs. probation on future criminal offending and employment? Data and sample: Administrative records on all felony convictions in Michigan from 2003-06 Records from courts, department of corrections, police, unemployment insurance agency Method: Quasi-experimental designs Using random assignment of judges to cases as instrumental variable Exploiting discontinuities in sentencing guidelines Guidelines restrict judges sentencing options based on (a) offense severity and (b) prior criminal record 8
My Role: Circuit Court Demographic Information Background research on operation of Michigan Circuit Court system Reading court documents Talking to judges and court officials Collecting data on judges ( part of new project sentencing disparities) Collecting data on judges from circuit court websites and Judgepedia Obtaining records from Michigan Supreme Court Administrative Office Biographical data on judges Circuit-level data on court processing 9
10 From 2003-2009 there were 289 judges in office 60% (n=173) were elected 40% (n=116) were appointed
Acknowledgements Jeffrey Morenoff, Ph.D. David Harding, Ph.D. MDOC & SCAO SRC Summer Internship Program 15
Urban Social and Built Environment and the Trajectories of Social Isolation: Findings from Detroit MI CHOICE Population Min Hee Kim (firstname.lastname@example.org) Social Environment and Health Program Sponsor: Philippa Clarke, Ph.D.
Internship Goals Analytic skills for multi-level data structure Explore the mechanisms through which neighborhood affects older adults health Engage in social environment and health scholarships Work and family balance 17
Background Why is social isolation important at later life? Staying at home, instead of admission to nursing home, has benefits at both individual and societal level Understanding social and built environment factors that affect social isolation is critical 18
19 Detroit older adults experienced rapid socioeconomic and structural decline in last decades
Research Question & the Focus How do neighborhood social and built environments explain the trajectories of social isolation, adjusting for socio-demographic and health factors? Focus on those who have unmet needs (i.e., Medicaid Waiver Program Recipients) in Central Detroit 20
Conceptual Model Individual Factors Socio-demographics & Health Baseline age, race, gender, education, housing type, marital status, being alone ADL and IADL limitation 21 Social Isolation Initial Status Social & Built Environment Street Conditions (% of Poor Streets in block) Social Disorder Index Residential Security Sign (% of Security Sign in block) Social Isolation Overtime
Methodology Analytic Methods Generalized Hierarchical Linear Modeling (HLM) Data 1) Michigan Minimum Data Set (MDS) for Home Care (2000-2008) followed every 90 days 2) Neighborhood Data using Systemic Social Observation (SSO) methods 22
SSO Data Neighborhood audit of all 4 streets in each clients residential block Using Google Street View (2007-2009) Indicators of built physical and social environment can be reliably assessed with a virtual audit instrument (Clarke, et al. (2010) Health and Place)
Social Isolation Social Isolation was measured as a dichotomous variable indicating whether clients level of participation in social, religious, occupational or other preferred activities declined As compared to the previous 180 days, as assessed by the case manager 24
Constructed Variables Difficulties with Activities of Daily Living (ADL) 7 items: Transfer, Walking, Dressing, Eating, Toilet, Grooming Bathing Individual item measured : 0 (independent) ~ 5 (activity did not occur) Difficulties with Instrumental Activities of Daily Living (IADL) 7 items: Meal, Housework, Money, Medications, Phone, Shop, Travel) Individual item measured: 0 (no difficulty) ~ 2 (great difficulties) Social Disorder Index (9 items) 1) Graffiti painted over; 2) Garbage, litter or broken glass; 3) Cigarette or cigar butts; 4) Empty beer or liquor bottles in streets, 5) Gang graffiti; 6) Other graffiti on buildings; 7) Abandoned car; 8) Condoms; 9) Drug related paraphernalia on the side walk 25
Individual Characteristics at Baseline (2000-2008) (N=1,009) (weighted) 26 Proportion or Mean (s.d.) Age 55 to 64 10.56 % Age 65 to 74 22.18 % Age 75 to 84 37.42 % Age 85+ 29.84 % Female 73.89 % African American 94.31 % <HS Education 49.90 % HS Education 44.39 % College and above 5.72 % House 57.00% Apartment 38.56 % Other Residential Type 4.44 % Married 20.77 % Not Staying Alone 55.40 % ADL Limitations 1.92 (1.21) IADL Limitations 1.55 (0.43
Neighborhood Characteristics at Baseline (2000-2008) (N=1,009) (weighted) Average % of poor street on the block 0.23 (s.d. 0.27) Average social disorder index 1.44 (s.d. 1.25 ) Average % of residential security sign in the block 0.02 (s.d. 0.78) 27 Proportion of Residential Security Sign Freq.Percent 092791.89% 0.1~0.25737.23% 0.590.89% Total1,006100%
Longitudinal Characteristics (2000-2008) (N=4,875) Average number of observation per person= 5.1 Weight generated based on the probability of retention Individual data was truncated at 3 years Average observations per neighborhood cluster 2.1 28
29 Table. Multilevel Logistic Regression Coefficients for Trajectories of Social Isolation: Detroit Minimum Data Set 2000-2008), Age 55+ (Obs=4,875) Uncond' Growth ModelGrowth Model + Socio-demographic Controls + Social and Built Environment Coef.(OR) Coef.(OR) Coef.(OR) Coef.(OR) Individual Fixed Effects Intercept 0.611.84 *** 0.34(1.40)-1.04(0.35)*-1.147(0.32)* Age 65 to 74 a 0.38(1.47)0.27(1.30)0.2518(1.29) Age 75 to 84 a 0.40(1.49)0.18(1.19)0.1458(1.16) Age 85+ a 0.11(1.11)-0.15(0.86)-0.176(0.84) Not Staying Alone -0.33(0.72)-0.323(0.72) ADL Limitations 0.07(1.07)0.0655(1.07) IADL Limitations 0.65(1.91)**0.6767(1.98)** Neighborhood Fixed Effects Average Street Condition -0.058(0.94) Social Disorder 0.0161(1.02) % Security Sign 3.1416(23.14)** TIME (Months) -0.020.98***-0.020.98*** 0.01 1.01 0.02 1.02 Variance Components6.69087 6.80317 7.506387.53534 ***P<.001; **P<.01, *P<.05, p<.10 §Note 1) We fixed the effects of time (i.e., constraining the random variance in time to zero), because there was no variability to be explained between persons. Still, we tested whether the trajectory slopes vary between persons by baseline characteristics. 2) Results are controlling for gender, race/ethnicity, housing type, and marital status
Discussions Practical implications Generalization to urban older adults population in poverty Some limitations to be further examined Methodological Implications Policy Implications 32
Thank you Special thanks to.. Philippa Clarke Ph.D., George Myers Ph.D., and 2012 Summer SRC Interns *Funding for the geocoding/SSO part of this project was provided through Grant number K01EH000286-01 (Clarke) from the Centers for Disease Control and Prevention (CDC) 33
Disclosure and Quality of Answers in Text and Voice Interviews on iPhones Monique Kelly Survey Methodology Program Sponsor: Fred Conrad, Ph.D.
Parent Study Examined Data quality (satisficing, disclosure, straightlining) Completion rates Respondent satisfaction Four existing or plausible survey modes that work through native apps on the iPhone 35
Experiment: 4 modes on iPhone Medium VoiceSMS Text Interviewing Agent HumanHuman voice (R speaks with I) Human text (R texts with I) AutomatedSpeech IVR (R speaks with system) Automated Text (R texts with system) 36
Items First, safe-to-talk question 32 Qs taken from major US social surveys and methodological studies E.g., Pew Internet & American Life Project Types of QS Yes/No Numerical Categorical Battery Items (series of Qs with same response options) 37
Respondents n = 642 iPhone users (age > 21) 158 to 165 randomly assigned to each mode Recruited from: Craigslist Facebook Google Ads Amazon Mechanical Turk Incentive $20 iTunes gift code 38
Summary Voice vs. Text Text produced higher data quality Greater disclosure, less satisficing, high satisfaction Human vs. Automated Automated interviews on a smartphone (in these modes) can lead to data at least as high in quality as data from human interviews in same modes No more satisficing than with human interviewers! More disclosure 39
Goals of Project To see how the interaction between R and the I agent differ across modes. How this explain differences in answers to same questions across modes. To understand interaction around disclosure of personal/sensitive information. 41
Example Research Questions Does more departure from the script reduce disclosure? automated interviewers never depart from script Do respondents exhibit less human-like communication (e.g. disfluencies) when interacting with automated speech system? 42
Rendering 43 Opened in Camtasia Then converted into an avi file PAMSS interface
Coding 45 Coding was done in a tool called Sequence Viewer.
Coding (continued) Respondent Codes Examples Answer question Partial answer Interviewer Codes Examples Ask question exactly as worded Ask question with wording change Questions Raised Possible Additions? 46
Relationship between sciptedness and disclosure. Whether I asks the question exactly as worded or not Comparison of Rs speech when I is human or automated. 47 Future Analyses
Conclusion Aim Interviewing agent effect on respondents answers. Project in early phases Three other modes to be transcribed, coded, and analyzed. Stay tuned for more! 48
Acknowledgements George Myers, Ph.D. Fred Conrad, Ph.D. Michael Schober Andrew Hupp Lloyd Hemingway Chan Zhang Mingnan Liu Chris Antoun The staff of Survey Methodology Program CMT 49
Understanding the Achievement Gap: Do Parent Expectations and School Climate Matter? 50 Adrian Gale, MSW University of Michigan Joint Program in Social Work and Developmental Psychology Sponsor: Toni Antonucci Ph.D. Life Course Development
Background The achievement gap. (Ferguson, 2003; Mandara et al., 2009; Woolley & Bowen, 2004) The reality of differential academic performance has implications for life outcomes. (Grogan-Kaylor and Woolley, 2010) Physical and Mental Health Marital and Parental Status Occupation and Income 51
Theoretical Framework: Bioecological Theory of Human Development 52 (Bronfenbrenner, 2004) Macro (e.g. national education policies) Exo (e.g. neighborhood, schools) Micro (e.g. parents, siblings)
Parent Expectations… Realistic beliefs about youth future achievement. Linked to child outcomes such as grades. (Yamamoto & Holloway, 2010) Differ by gender. (Wood et al. 2007; Wood et al, 2010) Found to be related to academic stereotypes and previous academic outcomes. (Ferguson, 2003) 53
School Climate… 54 Norms and expectations defined and perceived by individuals within the school. Related to childrens academic achievement. (Zullig, Koopman, Patton, & Ubbes, 2010) Multidimensional construct, typically studied from youths perspective. (Zullig, Koopman, Patton, & Ubbes, 2010) Parents perception of school climate shown to be related their parent aspirations for their children. (Spera, Wentzel, & Matto, 2009)
Research Questions RQ1: What is the impact of parent expectations and school climate on academic achievement? RQ2: Do parent perceptions of school climate moderate the effect of parent expectations on academic achievement? RQ3: What is the impact of student gender on parent expectations and school climate? 55
Description of Sample Wave 1 collected during 7 th grade (1991) N=1328 51% Male; 49% Female 66% Black; 34% White 57
Measures Parent Expectations Single item How far do you think (CHILD) will actually go in school? 9-point scale (1=8 th grade or less; 9=MD, JD or PhD) Parent perceptions of school climate 4-item scale (alpha = 0.84) Ex. Children generally feel that they belong 5-point scale (1=strongly disagree; 5=strongly agree) Academic Achievement 7 th grade GPA 5-point scale 58
Means (SD) of Variables Mean (SD)Range Parent Expectations6.8 (1.7)2-9 School Climate3.5 (0.6)1-5 Academic Achievement3.6 (0.9)1-5 59
RQ1: Impact on Academic Achievement Betab (SE)Sig. Level Parent Expectations.335.175 (.013)*** School Climate.093.131 (.035)*** N1224 R2R2.274*** 60 * p-value <.05; ** p<.01; *** p<.001 Models control for: race and gender.
RQ2: Interaction of Parent Expectations and School Climate Beta b (SE)Sig. Level Parent Expectations.335.175 (.013)*** School Climate.093.131 (.035)*** Parent Expectations X School Climate.234.026 (.019).172 N1224 Adjusted R-Square Main Effects Model.273*** Change in Adjusted R-Square Interaction Model.001 61 * p-value <.05; ** p<.01; *** p<.001 Models control for: race and gender.
RQ3: Impact of Gender on Parent Expectations and School Climate Parent Expectations Betab (SE)Sig. Level Gender (0=male:1=female).091.319 (.096)*** N1321 R2R2.01* School Climate Gender (0=male:1=female).014.017 (.036).625 N1303 R2R2.000 62 * p-value <.05; ** p<.01; *** p<.001Models control for: race.
Summary of Findings RQ1: Parent expectations and school climate were significant predictors of academic achievement. RQ2: No interaction of parent expectations and school climate on academic achievement. RQ3: Gender, significant predictor of parent expectations, but not related to school climate. 63
Discussion Parents role great. Expectations > Perceptions of School Climate Parent perception of school climate may not be as accurate because their interactions with school do not occur during class time. Parents expectations high for boys and girls. 64
Future Directions Examine relationships longitudinally to see their affect across time. Examine the three way interaction between school climate, parent expectations and gender on academic achievement. Examine the main and interactive effects of SES/race with gender, parent expectations, and school climate. 65
Income Inequality as a Predictor of Self-Rated Health Beth Simmert Ph.D. Student Department of Sociology Wayne State University Sponsors: Jessica Faul, Ph.D. Amanda Sonnega, Ph.D. Health and Retirement Study
SES/Health Gradient SES Rate of Chronic Diseases High Low
Background SES/Health Gradient Individual level mechanisms Increased access to health care Afford healthier foods Exercise Risky health behaviors Society level mechanisms
Background Theory of fundamental cause Persists across Time Age groups Racial groups Gender groups Source: Link and Phelan. 1995. Social Conditions as Fundamental Causes of Disease. Journal of Health and Social Behavior 35(Extra Issue):80–94.
Research Question Does income/wealth inequality explain differences in the SES/Health Gradient that are not accounted for by individual level measures of behavior and access to care for older Americans?
Independent Variables Gini Gender Education Age Wealth Income Median County Income Race Urbanicity
Health and Retirement Study (HRS) 2006 Health and Demographic data American Community Survey (ACS) 2005-2010 5-year estimates for Gini and county median income data Data
2006 HRS Data N=16,290Proportion Sex 59% Female 41% Male Race 84% Non-Hispanic White 16% Non-Hispanic Black Above/Below Median Income 60% Above 40% Below Location 76% Metro 24% Non-Metro Education 19% No degree 55% GED or High School degree 5% Some College 13% Bachelor Degree 9% Master or Professional Degree
2006 HRS Data N=16,290MeanMedian Age Group68.368 Wealth$562,633$205,750 Income$66,000$39,200
2006 ACS Data MeanMedianRange Gini Coefficient0.4470.4500.3320.601 Median County Income $51,254$47,500$20,081$115,574
Would you say your health is excellent, very good, good, fair, or poor? 1,844=Excellent 4,915=Very Good 5,030=Good 3,133=Fair 1,368=Poor Dichotomized into: Excellent, Very Good, Good=0 Fair, Poor=1 11,789=Excellent, Very Good, Good 4,501=Fair, Poor Dependent VariablesSelf rated Health
MethodLogistic Regression Model 1 Gini, sex, education, age Model 2 Wealth, income, county median income Model 3 Race Model 4 Level of urbanicity
Logistic regression for Self-Rated Health Model 1 OR 95%CI Model 2 OR 95%CI Model 3 OR 95%CI Model 4 OR 95%CI Gini coefficient 71.5 *** (25.5-200.6) 8.19 *** (2.72-24.63) 5.12 ** (1.63-16.07) 6.17 ** (1.85-20.62) Genderns *** Education0.63 *** (0.61-0.65) *** Age Group 1.13 *** (1.11-1.15) *** Wealth0.90 *** (0.89-0.92) *** Income0.75 *** (0.72-0.78) *** Median Income 0.72 *** (0.62-0.85) ***** Race 1.173 ** (1.06-1.30) ** Urbanicity ns * p<0.05, ** p<0.01, *** p<0.001
Logistic regression for Self-Rated Health: by Race Non-Hispanic White N=13,718 Model 3 OR 95%CI Gini coefficient25.61 *** (6.56-.99.99) Gender0.81 *** (0.74-0.88) Education0.75 *** (0.72-0.78) Age Group1.12 *** (1.10-1.14) Wealth0.888 *** (0.87-0.91) Income0.771 *** (0.74-0.81) Median Income0.788 * (0.64-0.98) Urbanicityns Non-Hispanic Black N=2,572 Model 3 OR 95%CI Gini coefficient0.039 * (0.003-0.516) Genderns Education0.80 *** (0.73-0.88) Age Group1.11 *** (1.06-1.16) Wealth0.94 *** (0.91-0.96) Income0.71 *** (0.65-0.78) Median Incomens Urbanicityns * p<0.05, ** p<0.01, *** p<0.001
Logistic regression for Self-Rated Health: by Gender Males N=6,721 Model 4 OR 95%CI Gini coefficient14.47 * (2.22-94.35) Education0.74 *** (0.70-0.78) Age Group1.13 *** (1.10-1.17) Wealth0.92 *** (0.90-0.95) Income0.75 *** (0.70-0.80) Median Incomens Racens Urbanicityns Females N=9,569 Model 4 OR 95%CI Gini coefficientns Education0.78 *** (0.82-1.18) Age Group1.10 *** (0.73-0.88) Wealth0.90 *** (1.06-1.16) Income0.75 *** (0.91-0.96) Median Income0.67 ** (0.65-0.78) Race1.23 ** (0.52-1.25) Urbanicityns * p<0.05, ** p<0.01, *** p<0.001
A society level measurement can be predictive of individual level health. The Gini coefficient is predictive of self-rated health. Higher inequality = higher probability of having poor self-rated health Gini coefficient remains significant after accounting for individual-level measures Greater effect in whites than blacks Greater effect in males than females Conclusions
More sub-group analysis High inequalityHigh heterogeneity Low inequalityLow heterogeneity/high income Low inequalityLow heterogeneity/low income Relative influence of others on measures of self-rated health. Implications for Future Research
Amanda Sonnega Jessica Faul George Myers ISR SAS Users Group Nicole, Adrian, MinHee, Monique, and Mara Janet Keller Michigan Square HRS faculty and staff Thank You!