SOC 20- Major & Minor Studies

Association Without Causation Studies Example

•  Things change or happen at the same time, no clear way to establish which one led to the other/ if one caused another at all 

• Ex: studies have found children that watch TV are more aggressive, without more information we can’t make this causation, maybe aggressive children are more into TV, maybe parents know more aggressive children can be behaved/ monitored by sticking them to a TV, we can’t conclude TV makes children more aggressive

• Ex: going to bed with shoes on is associated with waking up with a headache - this could be linked to going to sleep hungover with shoes on

• Ex: More Fire engines dispatched to the place the fire is more destructive, ex: we don’t know for sure - we send them because there is a fire

• Ex: study says eating more chocolate produces more nobel prize winners; the number of nobel laurreates is associated with annual chocolate consumption in country, for no particular reason countries produced more chocolate (association not causation)

• 2 things can occur at the same time, but not provide causation

• Important not to jump to conclusions

Ecological Validity (examples)

• Whatever we are studying replicate the real world, outcome of study mimics the real world

• EX: flight simulation not the same as real life / simulation is not going to be identical to reality, simulation ecological validity not very high

• Role-play in police training (don’t assume they translate into real world), police officer may perform differently with real world

• Ex: how you use language question varies from how you actually use language in social settings (for ex: if you were collecting information on the language of request making, asking people how they would respond to roommate leaving a mess in the kitchen, people would say they respond with “Can you clean up the kitchen?”, in reality, people may have hesitations/ may ask less directly given their social interactions with their roommate/ friend), people may make requests pointing/ with body language.

Experimental Design Examples

Ex:  think about identical plants, identical sunlight, the only manipulated variable is liquid used to water each plant (water,juice, study) 

• Dependent: height of the plant

• Independent: type of the liquid 

Experimental Design Across Groups Examples

Ex: Comparing coffees vs non coffee drinkers

Experimental Designs Across Conditions

Ex: Grouping gender and math/reading to test gender stereotypes (ACROSS CONDITIONS) 

• Stereotype Congruent (easy/fast) = brain jumps to these assumption, words sort faster (boy and math already sorted)

• Stereotype Incongruent (difficult/ slow) = words sort slower (boy is paired with reading)

• Have to choose left or right

• Unit of analysis is an individual 

• Across conditions (same people taking same set of group)

• Dependent variable: sorting speed, taken as proxy for stereotypes

Ex: punch with companion, punch without companion (measure happiness) =  across conditions (same unit of analysis) 

Ex: same plants one state (water) another state (orange juice)

Experimental Design

Strong in internal validity: control for cause and effect

Weak in external validity: one specific study

Weak in ecological validity: artificial set up (can be remedied by natural experiment)

STUDY BY AMADEU & DIVINE: ask whether implicit stereotyping or evaluative race bias are associated with sitting closer or further away from the belongings of someone they know is another race

ex - putting a bag down before people enter room, experiment already set up, trying to understand if people with a more stereotypical view will sit further away, people don’t know they are being observed

Cross-Sectional Design

Does money buy happiness?

• all american population

• how much money people make how happy they are

• they questioned multiple groups of people at one point in time

• Happiness: Dependent variable

• Money: Independent variable

• Internal Validity Weak: no control/ manipulation

• External Validity Strong: surveyed a large group of people

• Ecological Validity Weak(ISH): people can respond to surveys with some bias (humans bad at self-evaluation)

Longitudinal Ex:

Panel Study: Participants selected randomly (look at same people over multiple points)

Cohort study: participants have a common characteristic (born in 2000, graduated in 2025, get married on a certain day) 

• Looked at 4 students from disadvantaged background who were voted most likely to success, filmed over a ten year period, looked at their outcomes

Internal Validity: strong over a long period of time, you know before and after, units are the same

External Validity: weak; generalization not priority, sample attrition sample reduces, people loose motivation

Ecological Validity: weak(ish) depends on technique (observation higher, suvreys/ questionnaires) - weaker

Case Study Ex: Genie

The case of Genie - a critical case study for theories of language learning

• Linguists want to know how we learn language 

• Theory: To learn language, you have to be exposed to it by a certain age, past this point you can’t learn language anymore 

• Ex of Critical/ Extreme case - feral child: Genie discovered at 13 years old, kept in a basement in a cage, no interaction, tested theory of language, she could not produce speech 

• Tested whether exposure to language at a certain age is crucial for language development; answer was yes through this case study 

Internal Validity: very high (clear cause and effect)

External Validity: very low (not generalizable beyond study)

Ecological Validity: very high (observed real life outcomes)

Comparative Ex:

Applied to different units of analysis (exactly everything is the same, except units of analysis, establishes similarities/ differences) 

• Typically difference in unit of analyses is cross-cultural, cross-national, etc

• Ex: study amongst working class in NY/LA

From concepts to variable ex.

Concept = Academic ambition

Indicators

▪ Taking more than 12 units per quarter

▪ Caring about grades

▪ Participating in research

▪ Double majoring

▪ Taking graduate level courses

Variable = Amount of academic ambitiousness

Moss- Racusin Et Al. Summary

This study experimentally tested whether science faculty show gender bias when evaluating equally qualified students. Faculty rated identical application materials assigned either a male or female name. Results showed systematic bias favoring male students in competence, hiring, salary, and mentoring.

Moss - Racusin Et Al. Hypothesis

• Hypothesis 1: Faculty would rate male students as more competent, hireable, better paid, and more worthy of mentoring than identical female students.

  • Why: Prior research in social psychology shows persistent implicit gender bias and stereotypes portraying women as less competent in science, even among egalitarian (equitable) individuals. The authors extended this literature into academic science, where experimental evidence was lacking.

• Hypothesis 2: Faculty gender would not influence the bias (i.e., both male and female faculty would show similar bias).

  • Why: Previous studies indicate that implicit biases are culturally learned and widely shared, meaning even women may internalize stereotypes about women’s competence in STEM fields.

Moss- Racusin Et Al. Type of Research Design

The research design used in this study is an experiment, in which faculty evaluated identical application materials randomly assigned male or female names, allowing researchers to test for faculty bias while controlling for gender

• Manipulated Independent Variable (gender of the applicant)

• Random Assignment

• Control of all other variables (The application materials were identical in every way except for the name. That means any differences in evaluations can be attributed to gender, not other factors.)

• Measurement of outcomes (dependent variables):
  The researchers then measured things like competence ratings, hiring decisions, salary offers, and mentoring.

Validity - Moss Racusin Et. Al

• Internal Validity (High)

  • Only one variable manipulated: applicant gender

  • Identical application materials across conditions

  • Random assignment of participants

  • → Supports strong cause-and-effect conclusion (gender caused differences)

• Ecological Validity (Moderate)

  • Real science faculty as participants

  • Task (evaluating applications) reflects real-world behavior

  • Some artificial elements (study setting, not actual hiring decision)

  • → Fairly realistic, but not perfectly natural

• External Validity (Moderate)

  • Sample includes multiple universities and STEM fields

  • Improves generalizability within academic science

  • Limited by non-random sampling and U.S.-only context

Moss Racusin Et Al. - 3 Key Results

• Competence & Hireability Bias:
  Faculty rated male applicants as significantly more competent and more hireable than identical female applicants.

• Salary Difference:
  Male applicants were offered higher starting salaries (about $30,238 vs. $26,508).

• Mentoring Gap:
  Faculty were more willing to provide career mentoring to male students than female students.

Moss Racusin Et Al. Variables

• Independent Variable: Gender of applicant (male vs. female name)

• Dependent Variable: Faculty evaluations (competence, hireability, salary, mentoring)

Sampling Method

• Non-probability sampling: Faculty recruited from selected research-intensive universities

• Random assignment: Participants randomly assigned to applicant gender condition

Author’s Approach to Theory

Deductive approach

• The authors began with existing theories of implicit bias and tested specific hypotheses through an experiment.

Ontological orientation:

• The study leans more toward objectivism, not constructionism.

• Why: It treats gender bias as something that exists independently and can be measured objectively (e.g., through ratings of competence, salary).

WHO (Five) Well-Being Index

• Likert scale questions for five statements which indicate for well-being asking which is closest to how you have been feeling over the last 2 weeks. Higher numbers mean better well-being

• 5 (all of the time/ strongly agree)

• 4 (most of the time/ somewhat agree)

• etc.

Likert Scale

Strongly Agree (5)

Somewhat agree (4)

Uncertain (3)

Somewhat Disagree (2)

Strongly Disagree (1)

Kahneman and Deaton - High Income Improves Evaluation of life but not emotional well-being (Summary)

Summary
This study examines whether money increases happiness by distinguishing between emotional well-being (daily feelings) and life evaluation (overall life satisfaction). Using over 450,000 survey responses, the authors find that income increases life evaluation continuously, but emotional well-being only improves up to about $75,000/year, after which it plateaus

Kahneman and Deaton - High Income Improves Evaluation of life but not emotional well-being (Hypothesis)

Hypotheses (with reasoning)

• Hypothesis 1: Income is more strongly related to life evaluation than to emotional well-being.

  • Why: Prior research suggested that income affects how people think about their lives more than how they feel day-to-day.

• Hypothesis 2: Emotional well-being increases with income only up to a certain threshold, after which it levels off.

  • Why: Based on theories of adaptation and diminishing returns (e.g., Weber’s Law), increases in income have smaller psychological effects at higher levels.

Kahneman and Deaton - High Income Improves Evaluation of life but not emotional well-being (Research Design)

• Cross sectional

• Analysis of large-scale survey data (no manipulation of variables)

Kahneman and Deaton - High Income Improves Evaluation of life but not emotional well-being (Validity Assesment )

• Internal Validity (Moderate–Low)

  • No experimental manipulation → cannot establish causation

  • Many confounding variables (health, relationships, etc.)

  • Uses statistical controls, but causal claims are limited

• Ecological Validity (Low)

  • surveys subject to bias, surveys can not reflect reality

• External Validity (High)

  • Very large sample size (450,000+ participants)

  • Nationally representative survey methods

  • Findings generalize well to U.S. population (less certain globally)

Kahneman and Deaton - High Income Improves Evaluation of life but not emotional well-being (Main Findings)

• Money increases life satisfaction, but not daily happiness beyond ~$75K

• Emotional well-being depends more on factors like health, relationships, and loneliness

• Poverty worsens emotional suffering, especially in difficult life circumstances

• Supports the idea that “money buys life satisfaction, but not happiness”

Kahneman and Deaton - High Income Improves Evaluation of life but not emotional well-being (Variables)

Independent Variable:Income

Dependent Variable: emotional well-being/ life evaluation

Kahneman and Deaton - High Income Improves Evaluation of life but not emotional well-being (Sampling Method)

• Probability Sample: Selection is random, all members of population have an equal chance of being selected

Kahneman and Deaton - Author’s Approach to Theory

• Deductive

• Builds on existing theories and tests them with data

Univariate Analysis Ex

• Frquency of political views

• Histogram of arrival delays

• Frequency of units taken per quarter

Bivariate Analysis Ex

Examining two variables

Is income related to happiness?

• Gun ownership and gun deaths (scatterplot)

• Miles per gallon and vehicle weight (scatterplot)

• Hours of studying and grade point average

Multivariate Analysis Ex

Is alcohol consumption associated with GPA, independent of academic level?

• academic salary by years since degree

• academic salary by rank, sex, and years since degree

Telles & Lim (summary)

This study examines whether racial income inequality in Brazil changes depending on how race is measured: self-classification versus interviewer classification. Using a 1995 national survey, the authors find that interviewer classification produces larger white–nonwhite income gaps than self-classification. They argue that interviewer classification better captures racial discrimination because discrimination operates based on how others perceive race.

Telles & Lim (Hypothesis)

Hypothesis 1: Racial income inequality is greater when race is defined by interviewer classification than by self-classification.
Why: Discrimination depends on how others classify a person’s race, not how individuals identify themselves.

Hypothesis 2: Browns (mixed-race individuals) occupy an intermediate socioeconomic position between whites and blacks, but closer to blacks when using interviewer classification.
Why: Prior theory (e.g., Degler’s “mulatto escape hatch”) suggests browns may have intermediate status, but this depends on how race is socially perceived.

Hypothesis 3: Inconsistency between self- and interviewer classification systematically affects measured inequality.
Why: If individuals are classified differently depending on method, estimates of income gaps will shift.

Telles & Lim (Type of Research Design)

cross-sectional observational study

• Uses national survey data (Brazil, 1995)

• No experimental manipulation

• Compares measurement systems (self vs interviewer classification)

Telles & Lim (Validity Assesment)

Internal Validity (Moderate–Low)

• No random assignment or manipulation of race

• Cannot definitively prove discrimination causes income differences

• Possible omitted variables (social networks, discrimination history, regional effects)

Ecological Validity (High)

• Uses real-world labor market data

• Reflects actual social classification practices in Brazil

• Captures lived racial ambiguity and interaction-based classification

External Validity (Moderate–High)

• Large national sample (urban Brazil, ~4,000 cases)

• Likely generalizable to urban Brazil

• Less certain for rural Brazil or other countries, though authors suggest broader Latin American relevance

Telles & Limm Variables

Independent Variable:

• Race (self-classified vs interviewer-classified)

Dependent Variable:

• Income (ordinal categories transformed into log income via maximum likelihood estimation)

Control Variables:

• Age, age squared

• Sex

• Education (primary, secondary, college)

• Region (Northeast vs others)

• Urban size (large vs small cities)

Telles & Lim Sampling Method

Sampling Method

• National probability sample of urban Brazil (1995)

• Multi-stage random sampling:

  • municipalities → neighborhoods → streets → individuals

• Sample size: ~4,000 valid respondents (after exclusions)

• Only urban population included (age 16+)

• Stratified Random Sampling

Telles & Lim Main Findings

• Income inequality depends significantly on how race is measured

• Interviewer classification produces larger racial inequality estimates

• Self-classification tends to underestimate discrimination-linked inequality

• Racial categories in Brazil are fluid and socially constructed, not fixed

• “Race” operates as a social perception variable influencing economic outcomes

Telles & Limm Ontological Approach?

• Constructivist

• Racial categories in Brazil are fluid and socially constructed, not fixed

Sampling Error Ex

• Ex: 2016 presidential election (survey did not ask enough about education) 

• Pollsters did not ask for people’s education, non-educated people excluded from polls who favored Trump, voting patterns not accurately reflected 

• Non-response from Trump voters, if education was used as instrument in sampling, would have been more reflective of reality 

Campus Dining Ex

1. They randomly select names from an email list provided by the student union - error (sample frame problem- list is incomplete, tool to gain data incorrect) 

• Student union - not everyone is apart of that list, list is incomplete only undergrads with membership to union, not reflective of total population 

2. They randomly hand out survey leaflets only in the dorm cafeteria -bias 

• Accounting for only on-campus students, not commuters, also have a bias of students who are already there and are choosing/ may prefer food from cafeteria food over students who bring food back home 

3. They use the official university enrollment database, which includes all current undergrads, and randomly select students from that complete list to invite to the survey.  -none

Snowball Sample Ex

• Refer you to more people that belong to same social group you are interested in 

• (ex: people in illegal activity, no list online/documented, to understand these experiences you need to find referrals)

• (ex: anorexia, can’t assume they are listed with medical professionals/they may not be seeking help, referrals can help here) 

• (ex: disney world attendees, not a list easily available, may be easier to find somebody)

Stratified sampling

• You divide the population into groups

• Then you randomly select people from each group

• Trying to be somewhat representative

Quota sampling

• You divide into groups

• Then you non-randomly choose whoever is easiest to find until the quota is filled

• You can oversample something, doesn’t have to be proportionate, doesn’t have to be representative

Problem with Survey Research Ex (Trump Election)

• Most republicans say they doubt the 2020 election, but how many really mean it? 

• In early surveys done immediately after the election, a lot of Republicans responded as if they believed Trump had lost 

• When it comes to state or local elections, Republicans felt fairly good about how their votes were handled in 2020 

• Regardless of their true beliefs, Republicans said what they thought they should say in certain contexts; a report not about what they thought happened, but about their beliefs about what should have happened (answers vs reality) 

Reported Vs Actual Behavior

• Participants reported they would be less likely to honk their horn at a higher status car (luxury car) that was blocking traffic compared to a lower-status car (an old, beat-up car) 

• Later, when placed in an actual driving situation with a confederate’s car blocking traffic, participants honked equally at both higher-status and lower-status cars

Reported Vs Actual Behavior (Gender Talkativeness)

• A popular study from 2006 suggested a woman uses about 20,000 words per day while a man uses about 7,000

• But no systematic study , this study was false 

• Accurate study: EAR (electronically activated recorder)

• 396 participants (210 women/ 186 men) - not quota surveying here

• Women and men both use about 16,000 words per day (high individual variation)

• There is more variation within genders than between 

• Your response to this question is based on gender stereotype you are familiar with (reported vs actual behavior)

Reported Vs Actual Behavior (Church Attendance)

When asked in 1991 (Gallup polls), about 42% of adult Americans said they went to church or synagogue in the last 7 days

But in 1990s major denominations were not thriving and church service attendance appeared to be down 

• 36-37% of Protestant residents in Ohio and the Cincinnati area reported attending weekly

• But 19.6% Protestant attendance based on estimates of actual attendance in that area

• Actual Catholic attendance in US was 25% (rather than 51%)

Sampling Issue?

• Shouldnt be - we have a lot of data

Research Method Issue?

• Shouldnt be - we have good methods

• This gap is between reported vs actual behavior

A sturctured observation of medication rounds ex.

Aim

▪ Describe current practice in administering medication

in acute psychiatric unit

▪ Sample

▪ Convenience – 3 acute mental health wards

▪ Method

▪ Structured observation of 20 medication rounds

▪ Results

▪ 97% nurses showed warmth with good eye contact

▪ 46% initiated provision of information

▪ 35% inquired about patient health

▪ 17% inquired about medication problems

▪ 42% nurse responded to patient requests for info

Conversational Group Study

Research question

▪ Are there constraints on size and structure of

conversational groups?

Sample

Convenience – 3 public settings

▪ 802 “cliques”

Method

▪ Structured observation; 15-minute intervals

Results

▪ 54% of cliques involved 2 people

▪ 27% involved 3 people

▪ 13% involved 4 people

▪ < 6% involved 5-7 people

• on average groups are 3-4 people, typical size of conversation group found using this method 

Reactive Effects EX

Research question

▪ Do doctors change prescribing behavior during observational studies of medical

activity?

▪ Method

▪ Study of medical records at T1 + Observation study

▪ Study of medical records at T2 (1 year later)

▪ Results

▪ Inappropriate prescribing was significantly lower during observational study

(–29%!)

Stivers & Majid (2007) Pediatric Interaction Summary

This study analyzes how pediatricians decide whether to address questions to children or parents during medical visits, and how this depends on question type, child characteristics, and parent demographics. It argues that speaker selection in interaction reflects implicit judgments about competence and authority, and may contribute to healthcare inequality.

Stivers & Majid (2007) Pediatric Interaction Typoe of Study/ Research Design

• Quantitative observational study

• Structured naturalistic interaction analysis

• No manipulation of variables (non-experimental)

• Uses coded video recordings of real-world behavior

Stivers & Majid (2007) Pediatric Interaction Type Sampling Method

• Non-experimental convenience/field-based sampling of clinics

• Pediatricians and visits drawn from community practices in LA County

• Not nationally representative

• Clustered naturally around participating physicians and clinics

Stivers & Majid (2007) Pediatric Interaction Variables

Dependent Variable

• Question addressee selection

  • Parent (mother/father)

  • Child

Independent Variables

• Question content (medical vs social)

• Child age

• Parent race (e.g., Black, Latino)

• Parent education level

• Father present (yes/no)

• Doctor characteristics (race, gender)

• Child gender

• Interaction terms (e.g., race × education)

Stivers & Majid (2007) Pediatric Interaction Type Sampling Method

1. Addressee patterns

• 60% of questions → parent (mostly mother)

• 37% → child

2. Content effect

• Social questions → heavily directed to parents (+500% likelihood)

• Medical questions → more often directed to children

3. Child competence effect

• Older children more likely to be addressed (+22% per year)

4. Social inequality effects

• Parent identified as Black → significantly less child-directed questioning (–78%)

• Higher parent education reduces this disparity (interaction effect +32–56%)

5. No significant effects

• Doctor race

• Doctor gender

• Child gender

• Parent education alone

Stivers & Majid (2007) Pediatric Interaction Key Findings

1. Addressee patterns

• 60% of questions → parent (mostly mother)

• 37% → child

2. Content effect

• Social questions → heavily directed to parents (+500% likelihood)

• Medical questions → more often directed to children

3. Child competence effect

• Older children more likely to be addressed (+22% per year)

4. Social inequality effects

• Parent identified as Black → significantly less child-directed questioning (–78%)

• Higher parent education reduces this disparity (interaction effect +32–56%)

5. No significant effects

• Doctor race

• Doctor gender

• Child gender

• Parent education alone

Stivers & Majid (2007) Pediatric Interaction Validity

Internal Validity (Moderate–High)

• Strength: real-time behavioral data (not self-report)

• Controls for many confounders in regression model

• Limitation: observational design → cannot prove causation (only association)

---

Ecological Validity (Very High)

• Real pediatric consultations

• Natural interaction setting (no lab simulation)

• Captures authentic communication behavior

---

External Validity (Moderate)

• Limited geographic scope (LA County only)

• Small number of physicians (38)

• Likely not fully generalizable across all healthcare systems or countries