Foundations of Research Quasi-experimental & field studies This is a PowerPoint Show Click “slide show” to start it. Click through it by pressing any key. Focus & think about each point; do not just passively click. To print: Click “File” then “Print…”. Under “print what” click “handouts (6 slides per page)”. © Dr. David J. McKirnan, 2014 The University of Illinois Chicago McKirnanUIC@gmail.com Do not use or reproduce without permission Ongoing field studies show Baboons to learn and transmit “culture”. Click for NY Times article. 1 Foundations of Research Many important issues cannot be studied with true experiments. This module addresses four topics best addressed with quasi-experiments. Naturally occurring events / case studies Interventions within a single group Experiments with non-equivalent groups Time series designs & field observations 2 Foundations of Research 3 Quasi-experiments: how do we research… Naturally occurring events / case studies Interventions within a single group Experiments with non-equivalent groups Time series designs & field observations Group Naturally occurring event or social change Observe Foundations of Research Group 4 Naturally occurring events Naturally occurring event or social change Observe It is important to understand the consequences of phenomena that are naturally occurring. It is easy to think of examples: Social, cultural or political events War, 9/11-like events, immigration, torture. Policy changes; educational or civic policies, gay marriage Rapid economic change… Natural events Weather-related or other natural disasters Climate change… Behavioral events Crime victimization Childbirth… Foundations of Research Group 5 Naturally occurring events Naturally occurring event or social change Observe Research on naturally occurring events roughly falls into three categories: Simple description Predictors (correlates) of outcomes or consequences Potential interventions or policy changes 6 Naturally occurring events Foundations of Research Group Naturally occurring event or social change Observe Research on naturally occurring events… Simple description What is the nature – size, scope, intensity – of the event? It is ‘positive’ or ‘negative’? What are the consequences for those affected? Economic, social or cultural; are social structures damaged or enhanced? Individual; psychological adjustment or growth, physical harm, trauma Foundations of Research Group 7 Naturally occurring events Naturally occurring event or social change Observe Research on naturally occurring events roughly falls into three categories: Simple description Predictors (correlates) of outcomes What personal or social factors allow individuals or communities to benefit from ‘positive’ events? What personal or social factors create vulnerability to adverse outcomes? What confers resilience in the face of uncontrollable events; e.g. community support, economic resources Potential interventions or policy changes To prevent future occurrences To lessen adverse outcomes Foundations of Research Example: Child soldiers in Sierra Leone. 8 Description EXAMPLE During Sierra Leone’s civil war children as young as 7 were recruited as soldiers. • In this sample M age of captivity was 10, average duration = 5 years. • 88% witnessed violence or warfare • 28% killed a stranger or loved one. • 37% of females and 6% of males were victims of rape. • 7% experienced all three exposures. Most were resettled in their original villages or with relatives. Many returned child soldiers reported PTSD or other psychological symptoms • Anxiety, depression, hostility • Low confidence and prosocial attitudes. Betancourt, Theresa et al. (2010). Sierra Leone’s Former Child Soldiers: A Follow-Up Study of Psychosocial Adjustment and Community Reintegration. Child Development, 81 (4), Pp. 1077–1095. Click for article. Foundations of Research Child soldiers in Sierra Leone. Description Predictors / correlates of outcomes EXAMPLE Actions during the war were strongly related to negative outcomes • Wounding or killing another person • Witnessing violence • Victim of rape Family acceptance upon reentry was the key variable associated with resilience & recovery. • Community acceptance was less important short-term, more important long-term Family acceptance was weaker for those who were… • Older, male, engaged in or witnessed violence, or were raped. • Thus, acceptance was lower for those who needed it most. 9 Foundations of Research Child soldiers in Sierra Leone. 10 Note: “Predictors’ here are simply variables that Description correlate with the outcome. Predictors / correlates outcomes Causeofand effect can be difficult to EXAMPLE Actions during the war were strongly related to negative determine. outcomes Did family acceptance cause some • formerperson soldiers to have better outcomes? Wounding or killing another • Witnessing violence • Victim of rape Or, are soldiers who are doing better more easily accepted by their families? Family acceptance upon reentry was the key variable associated with resilience & recovery. • Community acceptance was less important short-term, more important long-term Family acceptance was weaker for those who were… • Older, male, engaged in or witnessed violence, or were raped. • Thus, acceptance was lower for those who needed it most. Foundations of Research Child soldiers in Sierra Leone. 11 EXAMPLE Description Predictors of outcomes Potential interventions or policy changes The predictor / correlation evidence has shown some paths toward interventions • International Rescue Committee has developed programs specifically targeting family & community integration • Programs for former child soldiers include Burundi, Democratic Republic of Congo, Ivory Coast, Liberia and Uganda. • UNICEF, the UN and other organizations are resettling soldiers There are only scattered data on the long-term effect of these programs. Here are interviews with Ishmael Beah, a child soldier in Sierra Leone who wrote a biography of his experiences. An excerpt. The complete interview. Foundations of Research Group 12 Naturally occurring events Naturally occurring event or social change Observe Let’s think about the study of naturally occurring events from a research perspective “Naturally Occurring” means events that are not controlled by a researcher. Group 13 Naturally occurring events Foundations of Research Naturally occurring event or social change No control over who is exposed to the event Possible control over who is selected for the research sample May compromise both internal and external validity. Event not controlled / manipulated. Not a true Independent Variable Often no control group Significant threat to internal validity Observe May or may not have control over measures. Archival measures – medical records, climate data, crime reports – may not assess exactly what the study needs. Survey or other posthoc measures can address hypotheses. Heuristic value of field studies: generating hypotheses for later experimental study or to confirm controlled data in a “real world” setting. Foundations of Research Examples of naturally occurring events: Descriptive data Events Outcomes Natural disaster / stressor The 1984 San Francisco earthquake Coping responses to a major stressor Trauma EXAMPLE Iraq / Afghanistan service PTSD & response to treatment Historical event 2007 – 2010 economic contraction Voting patterns Publicity / cultural event Negative results from The Women’s Health Initiative study of hormone replacement Health decisions 14 Foundations of Research Examples of naturally occurring events: Post-hoc ‘predictors’ (correlates) Events Outcomes Natural disaster / stressor The 1984 San Francisco earthquake Coping responses to a major stressor EXAMPLE Post-hoc surveys showed that: Trauma PTSDpredictor & response treatment Proximity to the event was the best oftostress Iraq / Afghanistan service reactions Historical event Previous psychiatric history was also predictive 2007 2010 economic – Those with anxiety or depressionVoting histories had a significantly patterns contraction stronger stress reaction. Social/ cultural support “buffered” the effect of the earthquake on Publicity event stress Negative results from The Health decisions People who Initiative perceivedstudy themselves to have strong social support Women’s Health networks fared better than those who did not. of hormone replacement 15 Foundations of Research Examples of naturally occurring events: Post-hoc ‘predictors’ (correlates) Events Outcomes Natural disaster / stressor The 1984 San Francisco earthquake Coping responses to a major stressor Trauma EXAMPLE Iraq / Afghanistan service PTSD & response to treatment Historical Post-hoc event studies have shown that: As–with earthquake data, both proximity to the event 2007 2010the economic Voting patterns contraction previous psychiatric history were strong predictors of PTSD. Publicity / cultural event Traumatic brain injury is a major co-factor in symptom Negative results from The Health decisions severity. Women’s Health Initiative study of Immediate ‘tela-treatment’ (web links from mental health hormone replacement professionals to the field) could moderate symptoms 16 Foundations of Research Examples of naturally occurring events: Post-hoc ‘predictors’ (correlates) EXAMPLE Descriptive & hypothesis-testing research has shown that: Outcomes Events Voters generally vote more conservative during times of economic stress. Natural disaster / stressor <15% of voters correctly estimate the huge gulf between The 1984 San Francisco Coping responses to a major the economic status of the top 10% vs. everyone else. earthquake stressor Voting rates among those who have lost their economic Trauma footing – ethnic minorities, the poor, young people – have PTSD & response to treatment gone down, notservice up. Iraq / Afghanistan Historical event 2007 – 2010 economic contraction Voting patterns Publicity / cultural event Negative results from The Women’s Health Initiative study of hormone replacement Health decisions 17 EXAMPLE Foundations of Research Examples of naturally occurring events: Post-hoc ‘predictors’ (correlates) Post-hoc surveys & prescription patterns have shown that: The great majority of post-menopausal women stopped Events Outcomes taking hormone replacement therapy altogether. Natural disaster / stressor Secondary analyses showed that; The San Francisco Coping responses to a major 1984 Estrogen was associated with a variety of health risks earthquake stressor Progesterone is protective for cancer and other diseases. Trauma & response to treatment Iraq Some researchers and cliniciansPTSD question whether the low / Afghanistan service absolute risk of estrogen justifies discontinuation in Historical event severe menopause. – 2010 economicmay identify which women can(not) 2007 Further research Voting patterns contraction tolerate estrogen. Publicity / cultural event Negative results from The Women’s Health Initiative study of hormone replacement Health decisions 18 Foundations of Research ‘One-shot’ survey study; Consumer Reports psychotherapy survey. 19 Seligman, M. E. P. (1995). The effectiveness of psychotherapy: The Consumer Reports study. American Psychologist, 50, 965-974. Despite experimental research showing that psychotherapy EXAMPLE is effective, there are few field studies of actual consumers. Many experimental studies from the 90s to today use small, carefully selected samples who are given ‘lab’-like therapy interventions. This is a major challenge to external validity. Seligman set out to answer: Does psychotherapy “work” from a consumer view? Who gets therapy / what does it consist of? Do consumer responses vary by type of treatment? This study remains one of the most often cited and controversial studies in the therapy outcome literature. Foundations of Research Consumer reports survey, 2 20 Research approach: One shot case study / survey Consumers given post-hoc survey of any therapy EXAMPLE experiences Sampling frame: Any therapy or psychological service user in the previous year We have little information about the general population of therapy users. Thus, we can never know if this sample is representative Sampling procedure: 4,100 Consumer reports readers responding to “in magazine” mail-back survey form Foundations of Research 21 Consumer reports survey, 3 Entire population of therapy clients U.S. Therapy clients Proportion of therapy clients who read Consumer Reports EXAMPLE Proportion of Consumer reports readers willing to return the survey CR Readers Key self-selection biases Returned survey Of the complete therapy client population, only a small % join Consumer Reports. Of therapy clients who join Consumer Reports, only a % were willing to complete a detailed survey. Neither of these percentages can be known, since we do not have data on people who did not return the survey. This source of bias may threated External Validity, since we cannot know how well these data generalize. The very large sample provides confidence, however. Foundations of Research Consumer reports survey, 3 Potential predictors of outcomes Gender, type & duration of treatment, medications. Negatives: EXAMPLE Cursory outcome measures: “satisfaction” & “helped with my problem” This may threaten Internal Validity, since we cannot ensure we are capturing the outcomes (e.g., change in mental health) we are interested in. Positives: Huge, national sample Anonymous, 3rd party data collection “Real world” assessment of product quality 22 Foundations of Research Consumer reports survey, 4. Descriptive findings: People generally felt EXAMPLE positive about their therapy experience. Analysis of predictors People who got more treatment did better. Demographic predictors (age, gender…had modest or no effects. For peoples’ presenting problem(s) all specialists did about the same. 23 Foundations of Research Consumer reports survey, 4. Analyses of predictors EXAMPLE For work and social domains: Marriage counselors were the least effective Physicians were only slightly better. 24 Foundations of Research Consumer reports survey, 5. Analyses of predictors EXAMPLE For personal domains (self-esteem…): Again marriage counselors were the least effective 25 Foundations of Research Consumer reports survey, 6. Bottom line Most therapy clients are satisfied EXAMPLE with their treatment. More treatment is better. Outcomes are not affected by age, gender, ethnicity… Mental health professionals get significantly better results than do marriage counselors or physicians 26 Foundations of Research Consumer reports survey, 7. Interpreting survey findings: Since these are simple descriptive EXAMPLE data we cannot determine why these outcomes occurred. Why might marriage counselors have done worse? Could marriage counselors be more poorly trained? Might less competent people enter the marriage counseling profession? Might marriages be really hard to fix? What data would allow you to test each of these alternate hypotheses? 27 Foundations of Research Naturally occurring events 28 Useful when: SUMMARY One An experiment is not possible There cannot be a control group “Pre-” measures not possible or practical Chief virtues: Describe naturally occurring or uncontrollable socially or politically important events Provides “real world” look at processes that are typically studied in experiments Archival data on potential predictors can help interpret the findings / “control” some alternate interpretations. Data may suggest interventions. Liability: lack of control group multiple threats to internal validity No pre-measure makes interpretation (e.g., of change…) difficult. Foundations of Research Quasi-experiments: Existing groups Naturally occurring events / case studies Single group interventions Experiments with non-equivalent groups Time series designs 29 30 One group pre- post-test Foundations of Research Group Observe1 Selected or convenience sample. Intervention or event Baseline Assessment May or may not have control over measures (e.g., surveys v. archival measures). Observe2 Outcome Assessment Typically controllable, but may be archival. Event or intervention May or may not be controllable by researcher, e.g., policy change. Uses: Educational & social environments Political or health policy change Not feasible to have a control group System-wide intervention / social change (school, public health campaign..) Group 31 One group pre- post-test Foundations of Research Observe1 Key design Observe 1 Intervention or event Observe2 feature: no control group. Observe2 Confound Threats to internal validity (confounds): History Historical / cultural events occur between baseline & follow-up. Maturation Individual maturation or growth occurs between baseline & follow-up. Reactive measures People respond to being measured or being a measured a second time. Statistical regression Extreme scores at baseline “regress” to a more moderate level over time. Mortality / drop-out People leave the experiment nonrandomly (i.e., for reasons that may affect the results…). Group 32 One group pre- post-test Foundations of Research Observe1 Intervention or event Observe2 Observe1 Confound Observe2 Threats to internal validity (confounds): History Historical / cultural events occur between baseline & follow-up. Maturation Individual maturation or growth occurs between baseline & follow-up. Reactive measures People respond to being measured or being a measured a second time. Statistical regression Extreme scores at baseline “regress” to a more moderate level over time. Mortality / drop-out People leave the experiment nonrandomly (i.e., for reasons that may affect the results…). Foundations of Research One group pre- post-test EXAMPLE Effects of HIV testing on sexual risk. Question: - Does HIV testing lead people to be sexually safer? Event: - Self-referred HIV testing & counseling Sampling frame: - Participants in testing centers Study structure: - Baseline retrospective interview at testing session - Follow-up interview 3 months later Quasi-controls: - Population characteristics as potential predictors of group differences in risk Outcomes: - Self-reports of sexual risk 33 Foundations of Research One group pre- post-test Effects of HIV testing on sexual risk. EXAMPLE Findings: - Significant shifts toward safety - Few demographic predictors of change Threats to internal validity - Self-selection into testing group - Mortality: non-random drop-out(?) - History: general shift in norms & behavior during study period - Instrument change; people answer more conservatively during a follow-up interviews 34 Foundations of Research One group pre- post-test EXAMPLE Effects of HIV testing on sexual risk. Thus a clinical follow-up design can provide important clues about a possible intervention… The constraints of this research approach make interpretation difficult. 35 SUMMARY Foundations of Research One group pre- post-test Virtues: Provides more systematic data on naturally occurring events Only possible design for system-wide intervention evaluations. Pre-measure allows researcher to interpret change & examine status of groups at baseline. Liability: Lack of a control group creates multiple threats to internal validity: History Maturation Statistical regression Reactive measures Mortality / drop-out 36 Foundations of Research 37 Quasi-experiments: Existing groups Naturally occurring events / case studies Single group interventions Experiments with non-equivalent groups Time series designs Group Observe1 Intervention or event Observe2 Group Observe1 Contrast group Observe2 Group1 Group2 38 Non-equivalent experimental designs Foundations of Research (No baseline) Groups are not equivalent at baseline, due to.. Self-selection into the experimental vs. control groups Non-random assignment into groups (e.g., first people to show up in experimental group…) Use of existing groups Participants are not blind to which condition they are in. Intervention or event Observe1 Contrast group Observe1 Assessments may or may not be controlled Survey or interviews Archival / existing data, e.g., clinic records, grades Intervention or event may or may not be controlled by the researcher; Existing program Experimental intervention Naturally occurring event 39 Non-equivalent experimental designs Foundations of Research Group Observe1 Intervention or event Observe2 Group Observe1 Contrast group Observe2 Non-equivalent groups Self-selection Non-random assignment Use of existing groups Participants not blind Intervention & Assessments often controlled by the researcher. Observation1 used to Assess equivalence of groups at baseline Assess change: the key outcome Test for threats to internal validity: Reactive measures History, mortality effects Regression effects Similar to true experimental design, except for non-equivalent groups Foundations of Research 40 True Quasi-experiments vs. Quasiexperiments True experiments Participant Participant Experimental Recruitment Assignment Procedures Experimental Treatment or Manipulation Results Group A Procedure A Treatment Outcome Group B Procedure A Control Outcome Sample Randomly selected from target population Convenience sample? Probability sample? Randomly assigned; groups identical at baseline. Self-selection? Non-random assignment? Existing groups? Non-blind? Procedures Complete Complete the Unbiased to groups same assignment control over control over exp. & IV. measures. for Participant and the experimenter blind control groups. Might selfselected or existing groups require different procedures? Naturally Archival occurring measures? event? (Not Existing a true IV) assessment? Foundations of Research Example of a formal Quasi-experiment EXAMPLE Safer sex intervention for drug using, risky MSM Multi-frame targeted sampling of gay/bisexual men Intervention group: 6 90-min. group clinical sessions Control group: 6 90-min. general group discussion sessions Men randomized at first group meeting Structured risk / attitude assessment at baseline, 3-, 6-, & 12month follow-ups. 41 Foundations of Research Quasi-experiment example Sample selection EXAMPLE Safer sex intervention for drug using, risky MSM We do not have a sampling frame for this sub-group of MSM. We necessarily use a multi-frame convenience sample. We ask men to call for enrollment, so there is self-selection into the sample altogether. However, men are randomly assigned to groups, to there is no self-selection there. 42 Foundations of Research Quasi-experiment example Sample selection Self-selection into the sample lessens external validity; the sample may be unlike the larger population. EXAMPLE Self-selection into groups would severely compromise internal validity; Participants may choose a group for reasons that affect thesex results, e.g., peoplefor really motivated change may Safer intervention drug using, to risky MSM select the intervention group… Since weahad random frame assignment to groups internal We do not have sampling for this sub-group of validity MSM was not compromised for this reason. We necessarily use a multi-frame convenience sample We ask men to call for enrollment, so there is self-selection into the sample altogether. However, men are randomly assigned to groups, to there is no self-selection there. 43 Foundations of Research Quasi-experiment example Group assignment EXAMPLE Safer sex intervention for drug using, risky MSM The groups are randomly assigned. (Not self-selected!) We try to convince participants that each arm of the intervention (control vs. treatment) are equal, but they still cannot be blind. Of course the interventionists cannot be blind. 44 Foundations of Research Quasi-experiment example Procedures EXAMPLE Safer sex intervention for drug using, risky MSM The procedures are highly standardized, so they are equivalent across group. All assessments are done with computer interviews, so the measurement procedures are also equivalent. 45 Foundations of Research Quasi-experiment example Treatment EXAMPLE Safer sex intervention for drug using, risky MSM We designed the experimental and control interventions, so we have complete control. (Different if we were assessing a “naturally occurring” therapy or health program.) 46 Foundations of Research Quasi-experiment example Bottom line EXAMPLE Safer sex intervention for drug using, risky MSM A randomized controlled trial of a behavioral intervention has both true- and quasi-experimental features. Groups cannot be perfectly equivalent Interpretation of findings have to take into account: Sampling methods Non-blind participants & interventionists 47 Foundations of Research True v. quasi-experimental designs 48 True experiments: Quasi-experiments: Emphasize internal validity Assess cause & effect (in relatively artificial environment) Test clear, a priori hypotheses Emphasize external validity Describe “real” / naturally occurring events Clear or exploratory hypotheses Participants randomly assigned to exp. or control groups Participants & experimenter Blind to assignment Non-equivalent groups Existing groups Non-random assignment Participants not blind Self-selection Full control may not be possible Control study procedures Manipulate independent variable May not be able to manipulate the independent variable Control procedures & measures Partial control of procedures & measures EXAMPLE Foundations of Research Example of a non-equivalent control group design; condom distribution Question: Does condom ed. & distribution: - increase safety - increase sexual activity Sampling frame: - Schools in New York & Chicago - Schools matched for SES, race, size Intervention: - Condom education & distribution in High School health classes Study structure: - NY = intervention schools, Chicago are contrast schools. - Baseline sexual health programming 9 mo. Follow-up Outcomes: - Clinical measures: STIs - Self-reports: sexual activity & safety 49 Foundations of Research Non-equivalent control group EXAMPLE Findings: NY (intervention) students; lower STI rate, safer sex NY and Chicago students; similar levels of sexual activity Thus; sexual health classes appeared to increase safety without increasing sexual activity. Internal validity?: Reactive measures; Study is not blind; NY students know they are the intervention group Non-equivalent groups: Possible differences between cities = unmeasured confounds 50 Foundations of Research Non-equivalent control group Most common quasi-experimental approach. Used where: Some form of control or contrast group is possible Groups cannot be equivalent: SUMMARY Virtue: Participants cannot be blind re: group assignment Random assignment not possible Must use existing groups Participants self-select into (or out of) groups. Study natural / “real world” interventions Contrast group lessens major threats to internal validity Liability: Using non-equivalent groups may introduce serious unmeasured confounds. 51 Foundations of Research Click Which of these is not a cause of nonequivalent groups? A. Self-selection B. Non-random assignment C. Use of existing groups D. Convenience sampling E. Participants not blind 52 Foundations of Research 53 Quasi-experiments: Existing groups Naturally occurring events / case studies Single group interventions Experiments with non-equivalent groups Time series designs Group Measure1 M2 M3 M4 Intervention or event M5 M6 … Foundations of Research 54 Simple time series design Group Measure1 M2 M3 M4 M5 M6… Group2 Measure1 M2 M3 M4 M5 M6… Group3 Measure1 M2 M3 M4 M5 M6… Examine / describe changes over time in one or more key variables. Describe or test hypotheses about group differences over time. Groups may be assigned, in a true experiment. … intervention groups with long-term follow-up Groups may be measured or naturally occurring. … age, gender or ethnic groups. Longer time-frame yields more valid & interpretable data. Foundations of Research Simple time series design; Are rates of childhood obesity lessening? 2003 2012 data Older kids (2 - 19): no change Toddlers appear to EXAMPLE be doing better. Supports effectiveness of recent infant programs. Longer time frame: 1999 2012 Older kids no still show no change Toddlers only look better because of a spike in 2003. Looking back to 1999 shows a flat line with lots of variance. http://www.nytimes.com/2014/04/15/health/obesity-studies-tell-two-stories-both-right.html? Original article: http://archpedi.jamanetwork.com/article.aspx?articleid=1856480 55 EXAMPLE Foundations of Research Simple time-series design; Childhood obesity. Time series designs Provide much more sensitive data than simple “one shot” measurements. But are still sensitive to the length and nature of their time frame. 56 Foundations of Research Group Interrupted time series design Measure1 M2 M3 M4 M5 Intervention or event Test effect of intervention or event on ongoing series of measurements. Intervention may be experimental or observed Policy shift, e.g., educational policy Uncontrolled event; e.g., 9/11/01, Media event Assessments may be experimental or archival Successive cross-sectional surveys Traffic data, clinic or crime reports, test scores M6 … 57 Foundations of Research Group Measure1 M2 Multiple baseline 58 Time series designs Demonstrate stable effect long-term crime rates disease prevalence economic performance… Or steady rate of change… M3 M4 M5 M6 … Intervention or event Hypothesis, tested by: Shift in stable rate after intervention. Increase / decrease in rate of change after intervention Foundations of Research Group 59 Time series designs Measure1 M2 M3 M4 M6 … M5 Intervention or event Multiple baseline Hypothesis; tested by: 4 Shift in stable rate after intervention 3.5 3 2.5 2 Intervention or event 1.5 1 0 1 2 3 4 5 6 7 8 9 10 Foundations of Research Group 60 Time series designs Measure1 M2 M3 M4 M6 … M5 Intervention or event Multiple baseline Hypothesis; tested by: 4 Shift in rate of change after intervention 3.5 3 2.5 2 Intervention or event 1.5 1 0 1 2 3 4 5 6 7 8 9 10 Foundations of Research Group Measure1 61 Time series designs M2 M3 M4 M5 M6 … Intervention or event Multiple baseline Hypothesis; Shift in stable rate. Increase / decrease rate of change Threats to internal validity: Sensitive to very local history Single group possibly prey to confound Advantage for internal validity Eliminates carryover effects of repeated measurement Tests maturation, history, reactive measurement, etc. Foundations of Example of interrupted time series: Research Shift in Baboon culture. 62 Question: Do baboon troops develop and transmit learned “culture”? EXAMPLE Baseline: Click for original article in PLOS Biology Long-term observational data on aggressiveness in a specific baboon troop. Intervention: Tuberculosis outbreak due to infected food. Dominant / aggressive males fed first Image: Shutterstock. are selectively infected are naturally culled from troop Naturally occurring event ongoing field study. Foundations of Research Baboon culture: findings 63 Outcome measures: Standardized indices of aggression & dominance. Core findings: EXAMPLE With dominant males gone, remaining males showed more cooperative behavior As new males entered the troop over time they were socialized into the cooperative culture Robert Sapolsky After a tuberculosis outbreak killed the most aggressive male baboons, the remaining members showed a greater willingness to foster a more patriotic spirit. Despite many new males over time, the troop remained cooperative for over 20 years. NY Times; Click image for article Click for an NYC Radio Lab description of the study; Is warfare innate in primates? Foundations of Research Example: Interrupted time series data The “Magic Johnson effect” on HIV testing Question: Does a celebrity or “role model” getting HIV EXAMPLE affect others…? Examining HIV testing rates before and after Magic’s announcement allows us to test the effect of a naturally occurring event. Magic Johnson announces that he is HIV-positive in 1991. Click for YouTube video of the announcement. 64 Foundations of Research Example: Interrupted time series data The “Magic Johnson effect” on HIV testing Data: Archival records of HIV tests reported to CDC, EXAMPLE collected monthly Data show stable baseline over multiple observations Timing of intervention precise relative to data collection Intervention: Magic reports infection on national TV. Uncontrollable, “naturally occurring” event Tests hypothesis re: modeling effects in health behavior Core Finding: Initial spike in testing rates, followed by leveling off at higher base rate. Initial increase expected Hypothesis tested by longer-term shift in testing rates 65 Foundations of Research 66 Example of time-series data: “Magic” / HIV effect. Time-series data showing shift in HIV testing after Magic’s announcement. Magic’s Announcement EXAMPLE Initial spike New, higher base rate Low & variable baserate of testing Multiple (monthly) measures. Tesoriero, J.M., Sorin, M.D., Burrows, K.A., LaChance-McCullough, M.L. (1995). Harnessing the heightened public awareness of celebrity HIV disclosures: “Magic” and “Cookie” Johnson and HIV testing. AIDS Education and Prevention, 232-250. Foundations of Research 67 Multiple time series data Group 1 Measure1 M2 M3 M4 M5 M6 … Group 2 Measure1 M2 M3 M4 M5 M6 … Groups typically formed by blocking variable measured post-hoc; Intervention or event Hypothesis; tested by an interaction of the blocking variable by the repeated measure: Health claims in NYC v. other cities post- 9/11/01 Younger v. older voting patterns post- Iraq invasion Heterosexual v. gay HIV testing rates post- Magic Johnson media event. Over repeated measurements, Does one group change while the other remains consistent? Do the groups change at different rates? Foundations of Research Blocking variables in the HIV testing data. 68 Core questions: EXAMPLE Heterosexuals and Ethnic minorities had low HIV testing rates Perceive HIV as a “white gay” problem? They may lack resources or venues for testing. Will having a prominent African-American Heterosexual disclose HIV+ status may change those perceptions? Hypotheses: Heterosexuals will respond more strongly than will gay/bisexual men. African-American and Latino men and women will respond most strongly. Foundations of Research Testing blocking variables: Gay / IDU data. 69 EXAMPLE Risky men & injection drug users: High baseline, high variability. Gay / bisexual men: less variable, but low baseline. Risky men & IDUs: slight increase, high variability. Gay & bisexual men: no change. Foundations of Research Testing blocking variables: Heterosexuals 70 EXAMPLE In contrast to gay / bisexual men or IDUs, heterosexual show an initially low baserate. Followed by a large spike after the announcement And a much higher new baseline. The hypothesis that heterosexuals would be more affected by the “Magic” announcement was supported by the interaction of Time x Sexual Orientation. Foundations of Research Testing blocking variables: Ethnic differences. 71 EXAMPLE African-Americans and Hispanics show low baseline and a high spike after the announcement Both groups go back toward their baseline shortly post-announcement. Foundations of Research Testing blocking variables: Ethnic differences. 72 EXAMPLE HIV testing among Whites was similar to African-Americans & Hispanics at baseline, They showed stable, much higher testing rate after Magic’s HIV announcement. EXAMPLE Foundations of Research Summary: Blocking variables in time series data A series of measures before & after an event allows us to clearly identify patterns of behavior, and to test group differences (via blocking variables). The hypothesis that ethnic groups would differ was supported by interaction of Time x the blocking variable of ethnicity (but in a direction that was not predicted: Whites showed more change). 73 Foundations of Research Time series designs: Summary Time series is most common with archival data: existing, standard records collected for other purposes. Used where: The hypothesis concerns changes in long-term trends Typically an experiment cannot be run Simple practicality or cost, e.g., health care issues Ethics; crime rates, rates of domestic violence, etc. The target events are not controllable. Virtue: Study natural / “real world” processes or interventions Blocking variables – comparing time trends across groups – allows us to test hypotheses. 74 Foundations of Research Click When studying naturally occurring events… A. The independent variable is often assessed after the event. B. The researcher has control over the study measures. C. It is possible to derive a random sample. D. Internal validity is typically high. 75 Foundations of Research Click Which of these is not a cause of nonequivalent groups? A. Self-selection B. Non-random assignment C. Use of existing groups D. Convenience sampling E. Participants not blind 76 Foundations of Research Exam issues Key terms: Slide 2: “true” v. quasi –experiments Threats to internal validity Basic forms of quasi-experiments Single shot Single group pre- post- test “Non-equivalent” two group designs: Self-selection (in or out [mortality]) Existing groups Non-blind Non- random assignment Interrupted time-series / group contrasts Virtues (external validity) and problems (internal validity) 77