Lecture 2: Field experiments
41 Questions
1 Views

Choose a study mode

Play Quiz
Study Flashcards
Spaced Repetition
Chat to Lesson

Podcast

Play an AI-generated podcast conversation about this lesson

Questions and Answers

In the context of field experiments on discrimination using name randomization on resumes, what critical assumption must hold true for the treatment effect $\gamma$ to be a valid estimate of discrimination?

  • The only systematic difference between the two groups of resumes (Black-sounding vs. White-sounding names) is the perceived ethnicity conveyed by the names, and this perception directly influences employer callback decisions without interacting with other resume attributes. (correct)
  • Employers must be completely unaware that a name manipulation is taking place and treat all applications as if they were genuinely representative of different racial demographics.
  • The distribution of unobserved characteristics affecting callback rates must be identical for both resumes with Black-sounding names and resumes with White-sounding names _before_ name assignment.
  • The selection bias arising from differential resume quality must be explicitly modeled and regressed out of the observed callback rate disparity.

According to the callback rate equations provided, $E[Y_{0i}] = \alpha_i$ and $E[Y_{1i}] = \alpha_i + \gamma N_i$, under what condition would a researcher be justified in concluding that $\gamma$ represents the causal effect of perceived race on callback rates, assuming $N_i = 1$ indicates treatment with a Black-sounding name?

  • Only when the effect of name perception ($\gamma$) is statistically insignificant, indicating no impact of perceived race on callback rates.
  • If the researcher can demonstrate that resumes with Black-sounding names are, on average, of lower quality than those with White-sounding names, thus necessitating a bias adjustment.
  • Given that $\alpha_i$ represents a random error term and, therefore, any observed difference in callback rates must be attributed to the name manipulation.
  • When $\alpha_i$ perfectly captures all non-racial determinants of callback rates, and there are no interactions between name-based perceptions and resume content. (correct)

In the context of the field experiments on discrimination, the equation $E[Y_{1i} - Y_{0i} | N_i = 1] = E[Y_{1i} - Y_{0i} | N_i = 1] + E[Y_{0i} | N_i = 1] - E[Y_{0i} | N_i = 0]$ is presented. What does the term $E[Y_{0i} | N_i = 1] - E[Y_{0i} | N_i = 0]$ represent, and why does it ideally vanish in a well-executed randomized experiment?

  • It denotes the systematic difference in potential callback rates that would exist even if the resumes with Black-sounding names had White-sounding names instead (or vice versa), and its disappearance signifies successful randomization. (correct)
  • It represents the component of callback rate disparity attributable to inherent differences in skill distribution across racial demographics, ideally vanishing due to statistical control.
  • It quantifies the selection bias arising from observable differences in resume characteristics and vanishes because resumes are meticulously matched on all attributes.
  • It captures the bias stemming from employers' subjective assessments of resume quality, eliminated by blinding employers to applicant names.

Consider a scenario where an audit study aims to measure gender discrimination in hiring. Researchers send matched resumes that are identical except for the applicant's name, where one set of names is identifiably female and the other set is identifiably male. However, unbeknownst to the researchers, employers in a specific industry sub-sector systematically favor candidates with extracurricular activities that are statistically more common among male applicants. What econometric challenge does this introduce, and how could researchers attempt to mitigate it?

<p>This introduces omitted variable bias. Researchers could mitigate it by collecting data on extracurricular activities and including them as control variables in their regression model to isolate the effect of gender. (D)</p> Signup and view all the answers

In Soetevent's (2005) field experiment on anonymity in giving within Dutch Baptist churches, what critical assumption underlies the causal inference regarding the effect of anonymity on monetary contributions, and how might unobservable violations of this assumption compromise the study's validity?

<p>The assumption that any external events or factors influencing overall church attendance and giving patterns are evenly distributed across both anonymous and non-anonymous offering settings, thereby mitigating potential confounding. (C)</p> Signup and view all the answers

In the context of experimental design, differentiate between the Hawthorne and John Henry effects by identifying the group primarily influenced and the nature of behavioral change exhibited.

<p>Hawthorne effect: Treatment group enhances performance due to awareness of being studied; John Henry effect: Comparison group enhances performance to counteract perceived disadvantage. (D)</p> Signup and view all the answers

A researcher conducts a study on a new teaching method, using volunteer participants. While internal validity is maintained, what primary concern arises regarding the generalizability of the study's findings to the broader student population?

<p>Volunteers may exhibit heightened motivation, leading to an overestimation of the method’s effectiveness. (C)</p> Signup and view all the answers

In the context of scaling up a small-scale educational voucher experiment to a nationwide program, what potential general equilibrium effect should policymakers anticipate that was not evident in the initial experiment?

<p>Increased competition among private schools, potentially leading to compromised educational standards, inflated tuition fees, and altered market dynamics. (C)</p> Signup and view all the answers

Consider a randomized controlled trial assessing the impact of a novel therapeutic intervention. If the control group, upon realizing their status, actively seeks supplementary treatments or modifies their health behaviors, which threat to external validity is most pertinent?

<p>John Henry Effect, manifesting as compensatory behavior among controls. (A)</p> Signup and view all the answers

A research team is evaluating the efficacy of a new organizational management strategy within a single department of a large corporation. Employees are aware of the study. Which of the following biases is most likely to confound the study's findings?

<p>Hawthorne effect, where awareness impacts performance. (C)</p> Signup and view all the answers

In the original Hawthorne factory studies, productivity increased regardless of whether lighting was increased or decreased. What key insight regarding experimental design does this illustrate concerning external validity?

<p>The potential for experimental conditions themselves to influence outcomes independently of the intended manipulation. (A)</p> Signup and view all the answers

An economist is evaluating a microfinance intervention in a small village. The intervention shows remarkable success in the controlled setting. However, policymakers hesitate to scale up the program nationwide. What represents the most significant concern regarding the potential for general equilibrium effects?

<p>Distorted local markets, resulting in the failure of informal credit systems. (B)</p> Signup and view all the answers

A pharmaceutical company conducts a clinical trial for a novel drug, where participants exhibit a noticeable improvement, irrespective of whether they receive the active drug or a placebo. Recognizing the influence of the Hawthorne effect, what refined methodological approach should be implemented to isolate and ascertain treatment-specific efficacy?

<p>Employing a double-blind design with a sufficiently large sample size. (A)</p> Signup and view all the answers

In the context of educational interventions, the John Henry effect manifests when students in the control group, aware of their status, exert extraordinary effort to outperform the treatment group. Which advanced statistical technique can researchers employ to disentangle the true intervention effect from the confounding influence of this compensatory behavior?

<p>Instrumental Variables (IV) regression using a valid instrument related to treatment assignment but not the outcome. (C)</p> Signup and view all the answers

A non-profit organization implements a new job training program in a small community. After one year, the program appears highly successful, with a significant increase in employment rates among participants. However, when policymakers attempt to replicate the program on a national scale, the observed employment gains are substantially smaller. Which of the following is the most plausible explanation for this discrepancy, considering general equilibrium effects?

<p>Increased competition for available jobs at the national level, caused by a large influx of newly trained workers, surpasses local demand, thus diminishing the overall impact of the program. (C)</p> Signup and view all the answers

In a randomized experiment with partial compliance, if policymakers are primarily interested in the impact of offering a treatment rather than the treatment's effect itself, which estimate is most relevant, considering real-world policy implementation?

<p>The Intention-To-Treat (ITT) estimate, reflecting the effect of the treatment offer, regardless of actual uptake, providing a pragmatic measure of policy impact. (D)</p> Signup and view all the answers

An experiment evaluating a new educational intervention exhibits differential attrition: students in the treatment group, who initially showed lower performance, are more likely to drop out. Which statistical method would BEST address the potential bias introduced by this non-random attrition, in order to estimate the true treatment effect?

<p>Inverse Probability Weighting (IPW), where participants are weighted by the inverse probability of remaining in the study, estimated using a logistic regression model based on baseline covariates and treatment assignment. (B)</p> Signup and view all the answers

In a clinical trial for a novel drug, researchers observe that attrition rates differ significantly between the treatment and control groups. Specifically, a higher proportion of participants in the treatment group discontinue the trial due to reported side effects, while more participants in the control group withdraw due to a lack of perceived benefit. Which statistical method would you use to obtain the LEAST biased treatment effect?

<p>Employing a multiple imputation (MI) technique, creating several plausible datasets that account for the uncertainty associated with the missing data, based on the reasons for attrition. (B)</p> Signup and view all the answers

Consider a large-scale social experiment offering job training to unemployed individuals. Participation is encouraged but not mandatory. Suppose an external economic shock disproportionately affects the control group, leading to higher unemployment rates specifically among those not receiving the job training. How would this event MOST likely impact the interpretation of the experiment's internal validity in estimating the causal effect of job training?

<p>It would compromise internal validity by introducing a confounding variable that differentially affects the control group's outcomes. (C)</p> Signup and view all the answers

In a randomized controlled trial (RCT) evaluating the effectiveness of a new mindfulness app on reducing anxiety, participants in the treatment group are given access to the app, while those in the control group receive standard care. However, a significant subset of participants in the treatment group do not consistently use the app, and some participants in the control group independently start using other mindfulness resources. Which of the following statistical approaches would MOST accurately estimate the effect of actually engaging with the mindfulness app intervention?

<p>Instrumental Variables (IV) estimation, using the original random assignment as an instrument for actual app usage. (D)</p> Signup and view all the answers

In a study evaluating the effectiveness of a new weight-loss drug, a researcher discovers that participants in the treatment group who experience early positive results are significantly more likely to remain in the study, while those who do not experience such results tend to drop out. Conversely, participants in the control group who perceive their weight as stable or decreasing are more likely to stay in the study, while those whose weight increases are more likely to drop out. This attrition process is MOST likely to:

<p>Introduce selection bias that systematically overestimates the true effectiveness of the weight-loss drug. (D)</p> Signup and view all the answers

A researcher conducts a randomized experiment to assess digital literacy training on employment outcomes, but finds that due to a concurrent government initiative promoting digital skills, many individuals in the control group independently seek and receive similar training. How does this MOST directly threaten the study's ability to accurately estimate the causal impact?

<p>By reducing the difference in digital literacy between the two groups, it dilutes the estimated effect size, potentially leading to a Type II error. (B)</p> Signup and view all the answers

In a randomized experiment evaluating a new therapy for depression, the research team discovers that a significant portion of participants in the treatment group are also receiving concurrent treatment from outside providers, which the researchers are unable to fully track. Simultaneously, some participants in the control group independently seek alternative treatments, creating a 'treatment contamination' effect. What is the MOST appropriate statistical approach to address this complex scenario and estimate the therapy's true effect?

<p>Apply instrumental variable (IV) estimation, using the initial random assignment as an instrument to account for the endogeneity arising from the treatment contamination. (D)</p> Signup and view all the answers

A researcher aims to assess how a new policy affects small businesses, but the policy implementation varies significantly across different regions due to local adaptations and pre-existing regulations. Some regions fully adopt the policy, others only partially implement it, and some barely change at all. How does this differential implementation MOST directly affect any assessment of causal impact?

<p>It complicates causal inference by introducing heterogeneity in the 'treatment', potentially leading to biased or inconsistent estimates. (B)</p> Signup and view all the answers

In the context of voucher programs and their impact on public schools, which of the following scenarios best exemplifies a general equilibrium effect that is virtually impossible to capture in a small-scale experiment, yet critically influences the large-scale program outcome?

<p>A significant, system-wide shift in public school resource allocation in response to competition from voucher programs, leading to both improved and worsened outcomes depending on the school's initial funding level and strategic response. (B)</p> Signup and view all the answers

Consider a field experiment designed to evaluate the efficacy of a novel educational intervention. The intervention shows promising results in a controlled setting. However, after scaling, the outcomes plateau. Which type of validity is most threatened by this plateau, and why?

<p>External validity, as the generalizability of the intervention's effects is compromised by contextual factors unique to the larger-scale implementation. (C)</p> Signup and view all the answers

What critical limitation is inherent in relying solely on earnings regressions with race dummy variables and control variables to estimate the effect of race on labor market outcomes, as highlighted by Bertrand and Mullainathan's work?

<p>The omitted variable bias introduced by unobserved factors correlated with both race and labor market outcomes compromises the validity of the estimated race effect. (D)</p> Signup and view all the answers

In the context of field experiments on labor market discrimination, what is the most significant advantage of the methodology employed by Bertrand and Mullainathan (sending out 'fake' job applications) compared to traditional audit studies utilizing actors?

<p>The use of a large number of 'fake' applications allows for the generation of statistically robust datasets, mitigating concerns about sample size limitations inherent in audit studies. (C)</p> Signup and view all the answers

How does Equation 2, $E[Y_{1i} - Y_{0i} | B_i = 1]$, address the fundamental problem of causal inference in the context of measuring labor market discrimination?

<p>It attempts to isolate the causal effect of being black ($B_i = 1$) on labor market outcomes by estimating the difference between potential outcomes ($Y_{1i}$ and $Y_{0i}$) for the same individual. (D)</p> Signup and view all the answers

What is the most critical assumption required for Equation 2, $E[Y_{1i} - Y_{0i} | B_i = 1]$, to provide an unbiased estimate of the treatment effect in measuring discrimination?

<p>That the assignment to the treatment group ($B_i = 1$) is as good as random, conditional on observed covariates. (A)</p> Signup and view all the answers

Consider a scenario where a researcher aims to replicate Bertrand and Mullainathan's field experiment in a different country with significantly different cultural norms and labor market structures. What adjustments to the experimental design would be most crucial to ensure the validity and relevance of the findings?

<p>Adapting the content of the resumes to reflect local educational qualifications, work experience norms, and common applicant characteristics relevant to the specific labor market. (D)</p> Signup and view all the answers

In the context of incidental experiments, such as draft lotteries or random assignment of judges, what potential threat to internal validity must be carefully considered when drawing causal inferences?

<p>Selection bias arising from non-compliance with the assigned treatment, leading to a violation of the intent-to-treat principle. (D)</p> Signup and view all the answers

A researcher aims to study the impact of a new microfinance program on poverty reduction using a field experiment. After two years, the researcher finds a statistically significant, but small, effect on income. However, qualitative data reveals that many participants used the loans primarily for consumption smoothing rather than investment. What type of validity is most directly challenged by this finding?

<p>Construct validity, because the intervention is not measuring the intended construct. (D)</p> Signup and view all the answers

What methodological challenge arises when attempting to extrapolate the findings from a lab experiment involving trust games to real-world economic behavior, and how might researchers mitigate this challenge?

<p>The artificiality of the lab environment may limit the external validity of the findings; researchers can incorporate elements of real-world context into the lab setting and conduct follow-up field experiments. (A)</p> Signup and view all the answers

In a perfectly randomized experiment, the difference-in-means estimator, formulated as $\frac{\sum_{i=1}^{n} D_{i} Y_{i}}{\sum_{i=1}^{n} D_{i}} - \frac{\sum_{i=1}^{n} (1 - D_{i}) Y_{i}}{\sum_{i=1}^{n} (1 - D_{i})}$, provides an unbiased estimate of which causal parameter under conditions of treatment effect homogeneity?

<p>The Average Treatment Effect (ATE) and the Average Treatment Effect on the Treated (ATET), predicated on the assumption of uniform treatment effects. (D)</p> Signup and view all the answers

Given the linear regression model $Y_i = \beta_0 + \beta_1 D_i + U_i$ employed in the analysis of a randomized experiment, which statement most accurately delineates the role of randomization in ensuring the validity of causal inferences derived from this regression?

<p>Randomization fundamentally ensures the statistical independence between the treatment assignment $D_i$ and the error term $U_i$, thereby satisfying the exogeneity criterion essential for interpreting $\beta_1$ as a causal effect. (D)</p> Signup and view all the answers

The provided text posits that randomization inherently fulfills the zero conditional mean assumption. Within the context of the regression model $Y_i = \beta_0 + \beta_1 D_i + U_i$, what is the most precise econometric interpretation of this fulfillment?

<p>$E[U_i | D_i] = E[U_i]$, indicating that the conditional expectation of unobserved factors is invariant to the treatment assignment status. (D)</p> Signup and view all the answers

Consider a randomized controlled trial conducted within an elite, specialized institution to assess a novel pedagogical intervention. The study demonstrates a statistically robust and substantial enhancement in student performance metrics. Evaluating this scenario through the lens of internal versus external validity, which statement is most tenable?

<p>The study's external validity is potentially constrained by the unique characteristics of the specialized institution, limiting the applicability of findings to broader educational settings. (D)</p> Signup and view all the answers

In a randomized controlled trial evaluating a vocational training program, significant partial compliance is observed: only 55% of those offered the training participate, and 15% of the control group independently access similar training resources. What is the most salient econometric implication of this partial compliance for the interpretation of the difference-in-means estimator?

<p>Partial compliance necessitates the application of instrumental variables estimation to recover the causal effect of actual treatment receipt, leveraging the randomized assignment as an instrument. (C)</p> Signup and view all the answers

Within an encouragement design, individuals are randomly provided with a voucher intended to facilitate access to preventative healthcare services; however, the utilization of these services remains discretionary. In such a design framework, which causal parameter is most directly and consistently estimated by the difference-in-means estimator when comparing outcomes between the voucher-offered group and the voucher-not-offered group?

<p>The Intent-to-Treat Effect (ITT) of the voucher offer on subsequent healthcare utilization and related health outcomes. (A)</p> Signup and view all the answers

A researcher initiates an experiment to examine the efficacy of an advanced neurocognitive training regimen on fluid intelligence. Participants are randomized to either receive an invitation to enroll in the training or not. The training is delivered exclusively via a high-bandwidth online platform, and participant access to reliable internet infrastructure varies considerably. Furthermore, a subset of the control group independently adopts comparable cognitive training methodologies available through open-access online resources. Which of the following statements most accurately characterizes the predominant threats to achieving valid causal inference and the most appropriate analytical strategy in this complex scenario?

<p>The principal challenges are partial compliance and potential attrition bias correlated with internet accessibility, necessitating an Intent-to-Treat (ITT) analysis, potentially supplemented by instrumental variables estimation for deeper insights. (C)</p> Signup and view all the answers

Flashcards

ATE = ATET (with randomization)

With randomization, the Average Treatment Effect (ATE) equals the Average Treatment Effect on the Treated (ATET).

Difference-in-means estimator

Estimating ATE by comparing the average outcomes of the treatment group to the average outcomes of the control group.

Regression with Randomization

A regression model where the treatment status is included as a predictor, and randomization ensures the treatment is independent of other unobserved factors.

Randomization and Unobserved Factors

Ensures the treatment status is not systematically related to unobserved factors.

Signup and view all the flashcards

Internal Validity

The extent to which an experiment provides a valid estimate of the causal effect for the population being studied.

Signup and view all the flashcards

External Validity

The extent to which the estimated causal effect from an experiment can be generalized to other populations, settings, or treatments.

Signup and view all the flashcards

Partial Compliance

Occurs when individuals do not adhere to their assigned treatment (either by not taking it when assigned, or by taking it when not assigned).

Signup and view all the flashcards

Intention-to-Treat (ITT) Estimates

Estimates resulting from partial compliance with the treatment in an experiment.

Signup and view all the flashcards

Z vs. D in ITT

Variable randomly assigned in an experiment (Z) may differ from the actual treatment (D) received.

Signup and view all the flashcards

ITT Relevance

The causal effect of the randomly assigned variable (Z). Useful when policies don't reach everyone.

Signup and view all the flashcards

Attrition

A threat to internal validity where participants drop out of the experiment.

Signup and view all the flashcards

Non-random Attrition

Attrition that is not random and is related to the treatment.

Signup and view all the flashcards

Attrition and Optimizing Behavior

Participants may drop out if they gain little from the treatment.

Signup and view all the flashcards

Monitoring Dropouts

Monitoring and comparing dropouts in treatment and control groups.

Signup and view all the flashcards

Nonrepresentative Treatments

A threat to external validity where treatments do not generalize to other settings.

Signup and view all the flashcards

Leap of Faith

Assuming a treatment will have the same impact across different places and times.

Signup and view all the flashcards

Hawthorne and John Henry effects

When the experimental setting itself alters behavior, impacting external validity.

Signup and view all the flashcards

Hawthorne effect

Changes in behavior in the treatment group due to being studied.

Signup and view all the flashcards

John Henry effect

Changes in behavior in the comparison group due to being studied.

Signup and view all the flashcards

General equilibrium effects

Occurs when experiment’s effects wouldn't hold if scaled up to a larger program due to changes in the broader economic environment.

Signup and view all the flashcards

Non-generalizable sample

A group of individuals selected for a study that may not accurately represent the broader population

Signup and view all the flashcards

Being experimentally measured

The change in a subject's behavior simply from being studied, not from any specific manipulation.

Signup and view all the flashcards

Placebo effect (in experiments)

The effect of a treatment that includes unintended psychological effects.

Signup and view all the flashcards

Awareness of observation

Experiment participants are aware they are being observed.

Signup and view all the flashcards

Volunteer bias

Volunteers may be more motivated, leading to magnified treatment effects.

Signup and view all the flashcards

Field Experiment

A research method where researchers manipulate variables in a real-world setting to study causal relationships.

Signup and view all the flashcards

General Equilibrium Effects (Vouchers)

Effects observed only on a large scale, influencing program outcomes via public school pressure shifts or student/parental selection.

Signup and view all the flashcards

Lab Experiments in Economics

Experiments where researchers randomly assign characteristics within a controlled environment.

Signup and view all the flashcards

Manipulating Perceptions of Race

The perception of race is manipulated rather than race itself in experiments, often through names.

Signup and view all the flashcards

Discrimination Measurement

Differential rates of callbacks from employers based on the perceived race (inferred form names) of job applicants.

Signup and view all the flashcards

Field Experiments

Experiments conducted in real-world conditions.

Signup and view all the flashcards

Incidental Experiments

Experiments arising from naturally occurring events or policies that create random assignment.

Signup and view all the flashcards

Bias Term (in Discrimination)

Differences in potential callbacks for resumes based on perceived race, absent randomization.

Signup and view all the flashcards

Resume Equivalence

Ensuring resumes are identical except for names eliminates inherent differences and ensures a fair evaluation

Signup and view all the flashcards

Bertrand & Mullainathan (2004) Method

Sending out numerous fake job applications to real jobs to assess hiring discrimination.

Signup and view all the flashcards

Treatment Effect (Discrimination)

Estimating the impact of being Black (Bi = 1) versus White (Bi = 0) on labor market outcomes.

Signup and view all the flashcards

Early Regression Approach

Using earnings regressions with a race dummy variable and control variables to estimate discrimination.

Signup and view all the flashcards

Audit Studies

Employing actors of different races in simulated interviews to measure discriminatory behavior.

Signup and view all the flashcards

Audit Studies: Limitations

Challenges arise from unobserved variables and ensuring actors behave identically.

Signup and view all the flashcards

Bertrand & Mullainathan Solution

Sending many fake applications solves the small numbers/behavior problem, and allows for a measure of discrimination in real-world hiring.

Signup and view all the flashcards

Study Notes

Analyzing Data from Experiments

  • Randomization implies ATE (Average Treatment Effect) = ATET (Average Treatment Effect on the Treated)
  • ATET is calculated as the expected difference in outcomes between the treated and control groups

Regression and Randomization

  • Regression model can be specified as Yi = Bo + B₁Di + Ui
  • Randomization ensures that Di is independently distributed of the unobserved factors in Ui
  • Treatment status is not systematically related to any unobserved factors, fulfilling the zero conditional mean assumption

Internal and External Validity of Experiments

  • Internal validity assesses if the experiment provides an accurate estimate of the causal effect in the population
  • External validity is the extent to which the causal effect can be generalized to other populations, economic settings, or related treatments
  • A trade-off exists between internal and external validity

Threats to Internal Validity: Partial Compliance

  • Partial compliance occurs when only a fraction of those offered the treatment take it up
  • It also occurs when the comparison group receives the treatment
  • Experiments that don't intend to treat everyone are called encouragement designs
  • Randomization affects the probability of exposure to treatment
  • With partial compliance, fine resulting estimates are intention-to-treat (ITT) estimates
  • Z is variable randomly assigned while D is the actual treatment when ITT-estimates work
  • E[Yį |Zi = 1] - E[Yį |Z₁ = 0] equals the causal effect of Z, but not the same as the effect of the treatment, D
  • ITT estimates may be of interest when policies won't reach everyone
  • Partial compliance can compromise external validity
  • Randomization can be used as an instrumental variable to recover the Local Average Treatment Effect (LATE)

Internal Validity: Attrition

  • Attrition, where individuals drop out, threatens internal validity
  • Random drop-out is not a problem, but reduces statistical power
  • Non-random drop-out related to treatment is a problem
  • Optimizing behavior can cause dropout, with those gaining little dropping out more
  • Attrition rates may be equal between treatment and control, but produced differently

External Validity

  • "Treatments" and policies will not always have the same impact in other places and times
  • Individuals may not be a random sample of the population
  • Volunteers in medical studies exemplify non-generalizable samples
  • Volunteers may be more motivated, leading to a greater treatment effect

External Validity: Hawthorne and John Henry Effects

  • Hawthorne and John Henry effects happen when being in an experiment changes behavior
  • Hawthorne effects refer to changes in behavior among the treatment group
  • John Henry effects refer to changes in behavior among the comparison group
  • The study of light manipulation in the Hawthorne factories in the 1920s, found increase in productivity no matter the type of light
  • A legendary American steel worker in the 1870s, upon learning his output was being compared with that of a steam drill, pushed extremely hard to outperform the machine that caused his eventual death
  • This raises concern as Hawthorne and Henry effects might not be the same in other settings

External Validity: General Equilibrium Effects

  • Experiments often fail to reflect general equilibrium effects that occur when scaled up
  • A small-scale experiment may not reflect the effects of a large--scale programs
  • Vouchers may increase pressure on public schools, improving performance
  • Vouchers may pull motivated children/parents out of public schools, potentially reducing pressure

Examples of Experiments in Economics

  • Experiments are increasingly common in economics
  • Lab experiments randomly assign the characteristics of the game, e.g., trust games
  • Field experiments take experiments to the field
  • Incidental experiments, e.g., draft lotteries, random assignment of judges

Example of Field Experiments: Discrimination

  • Measuring discrimination on the labor market involves estimating the "effect" of race
  • Early evidence used earnings regressions with a dummy variable for race
  • Bertrand and Mullainathan (2004) examined if Emily and Greg were more employable than Lakisha and Jamal
  • They manipulated perceptions of race by assigning ethnic names to job applications
  • This study measured discrimination by comparing callback rates based on name sound
  • The probability of a callback can be written as E [Yoi] = ai if the name is white-sounding(Yoi)
  • The probability of a callback can be written as E[Y1i] = a¡ + γN₁ if the name is black-sounding(Y₁₁)
  • For resume i, if N₁ = 1, the treatment effect becomes E[Y1i - Yoi Ni = 1] = a¡ + γΝ¡ - α¡ = γ
  • The difference in potential call-backs for resumes with a black and white-sounding names equates to the Bias term
  • Since resumes are identical except for names, E [Yoi Ni = 1] = E[Yoi Ni = 0] where the bias term cancels out
  • Differences in callback probability are due to random name assignment

Field Experiments on Anonymity

  • Anonymity in Giving in a Natural Context, written by Soetevent, A. (2005), was measured using the collection bag and baskets
  • The causal question asked how anonymity affects giving contributions to good causes
  • Anonymity of giving to offerings were randomized in 30 Baptist churches in the Netherlands
  • In this study, using baskets initially increased contributions by 10% but this effect decreased over the experimental period
  • People tended to switch to giving larger coins when baskets were used, even when the total amount was the same
  • Receiving approval from others play an important role in giving, and the anonymity removal triggers this

Incidental Experiments: Policymakers

  • Chattopadhyay and Duflo (2004) created an incidental experiment to study women as policy makers
  • The causal question asked if female politicians implement different policies than male ones
  • Finding the answer includes men and women having different preferences
  • Also finding the answer means the identity of policymaker must affect outcomes
  • In these councils only women were elected to the position of head
  • A sample of villages were surveyed in two districts: Birbum in West Bengal and Udaipur in Rajasthan
  • Villages were selected to be reserved for women at random
  • Differences in investment can be contributed to villages reserved for women when chosen at random
  • Women invest more in infrastructure that is directly relevant to the needs of their own genders

Summary of Social Experiments

  • Randomized trial is the "gold standard" to establish causality
  • A "true" experiment can evaluate non-experimental experiments
  • Starting point may not be random, a good starting point is to think of the perfect experiment to run
  • Then one must come is close as possible to what would have been the perfect experiment, using non-experimental data

Studying That Suits You

Use AI to generate personalized quizzes and flashcards to suit your learning preferences.

Quiz Team

Related Documents

Description

In discrimination studies using field experiments, a key assumption is that names are randomly assigned. Random assignment ensures that any observed differences in outcomes are due to name and not other factors. This allows for a valid estimate of discrimination.

More Like This

Use Quizgecko on...
Browser
Browser