Questions & explanations
1. Give an example where forward induction eliminates an equilibrium that backward induction does not.
Consider a game where player 1 first chooses between an outside option and entering a subgame. In the subgame, player 2 chooses between a fight and yield. Suppose the outside option gives a moderate payoff, while in the subgame, if player 2 fights, both get low, and if yields, both get a higher payoff. Backward induction may select the equilibrium where player 1 enters and player 2 yields. But forward induction: if player 1 chooses the outside option, player 2 might think player 1 feared a fight, but that is not rational. Actually, suppose there is another equilibrium where player 1 stays out and player 2 threatens to fight. Forward induction eliminates the threat because if player 1 enters, player 2 would deviate to yield, so the threat is not credible. Thus forward induction selects the entry equilibrium.
2. What is a limitation of survey experiments regarding external validity?
External validity means whether the results apply to real-world settings. Survey experiments often use volunteer respondents (like from online panels) who might not represent the general population. The artificial setting—reading a short text or seeing an image online—may not mirror how people encounter information in daily life. Also, the effects measured are usually immediate, but real-world influences might wear off or be strengthened by repetition. For example, a survey experiment might show that a negative ad reduces support by 5% right after viewing, but in an actual campaign, the same ad might be seen many times and compete with other messages. So the true effect could be different. Researchers must be careful not to overgeneralize survey experiment findings.
3. Give an example of a conjoint experiment in a survey and what it measures.
A conjoint experiment shows respondents several profiles of, say, political candidates, with attributes like party, age, and policy positions. Each profile is randomly varied. For instance, one profile might be a 45-year-old Democrat who supports tax cuts. Another might be a 55-year-old Republican who supports health care reform. Respondents choose which candidate they prefer. Because attributes are randomly assigned, we can calculate the average causal effect of each attribute on choice. For example, we can see how much more likely a candidate is chosen if they are a Democrat versus Republican, or if they support tax cuts versus health care. This method measures trade-offs and preferences without asking direct questions that might be biased.
4. Compare a field experiment and an observational study for measuring the effect of campaign calls on voting.
In a field experiment, researchers randomly assign some registered voters to receive a phone call encouraging them to vote, and others to get no call. Then they compare turnout rates using official records. This is the gold standard because random assignment ensures the only difference is the call. An observational study would just compare people who answered a call to those who did not, but those who answer might be more civically engaged anyway, leading to a biased estimate. The field experiment eliminates that bias. However, field experiments are expensive and can only test one message at a time. Observational studies can use large existing data sets, but they require complex statistical controls to try to mimic random assignment.
5. How does a survey experiment compare to a field experiment in studying media effects?
A survey experiment takes place inside a survey, where researchers control the exact message and measure immediate reactions. It is cheap, fast, and can test many variations. A field experiment happens in real life, like showing a political ad on TV to randomly selected households and later checking their voting behavior. Field experiments have higher realism because people are not aware they are in a study. However, they are expensive and harder to control other exposures. Survey experiments can show strong internal validity—we know the cause—but their results may not apply outside the artificial setting. Field experiments have better external validity but less control. Both are useful, and together they provide stronger evidence.
6. How can we estimate the causal effect of an institution like a democracy on economic growth?
Estimating the effect of democracy on growth is tricky because democracies differ from non-democracies in many ways. One approach is to use a natural experiment, such as when some countries became democratic due to historical accidents not related to growth. For example, researchers have looked at countries where democracy was randomly gained after a leader's death. They compare growth in countries that became democratic to those that did not, using statistical methods to account for other factors. Another method is instrumental variables, where we find a variable that affects democracy but not growth directly. Studies using these approaches often find that democracy has a modest positive effect on growth, but the results vary.
7. What does causal inference mean for media effects?
Causal inference for media effects asks whether exposure to a news story, ad, or social media post actually changes people's opinions or behavior. For example, does watching a negative campaign ad cause people to dislike a candidate? Because people choose what media to consume, those who watch negative ads might already dislike the candidate. To find a true cause, we need to compare what happens to people who are randomly exposed to the ad versus those who are not. This can be done with experiments, like showing one group the ad and another a neutral video. The difference in attitudes after exposure is the causal effect. Without such methods, we might incorrectly think media is powerful when it only reinforces existing views.
8. What is causal evaluation of development interventions?
Causal evaluation asks whether a development program, like building a school or giving cash to the poor, actually improves lives. For example, does providing free school meals cause children to attend more? To answer, we need to compare what happened with the program to what would have happened without it. This is hard because we cannot see both worlds. The best way is a randomized controlled trial (RCT), where some villages get the program and others do not, chosen by lottery. Then any difference in outcomes between the groups is due to the program. Without such methods, we might waste money on programs that seem to work but actually do not cause change. Causal evaluation helps governments and aid agencies spend effectively.
9. What is causal inference in elections?
Causal inference in elections means figuring out whether a campaign action or event actually changed voters' choices or turnout. For example, does a TV ad cause more people to vote for a candidate? Or does a get-out-the-vote phone call cause people to actually vote? Because voters are not randomly assigned to see ads or get calls, we cannot simply compare those who saw and did not see. Other factors, like being interested in politics, might explain both seeing ads and voting. Researchers use experiments or statistical methods to isolate the cause. A common method is to randomly assign some voters to receive a message and compare their behavior to others who did not. This gives a true causal estimate of the campaign's effect.
10. Give an example of a natural experiment used to study campaign effects on turnout.
A natural experiment happens when something outside the campaign creates random variation in exposure. For instance, some states hold elections on the same day as a popular presidential race, while others hold separate off-cycle elections. The timing is not decided by campaigns, so it acts like a random assignment. Researchers can compare turnout in places with on-cycle versus off-cycle local elections. The difference in turnout is likely caused by the presence of a high-profile race, not by voter interest. Another example: bad weather on election day reduces turnout, and since weather is random, the effect of rain on turnout can be estimated. These natural experiments help us understand what really drives people to vote.
11. Why is random assignment important in a survey experiment?
Random assignment makes the groups similar in all ways except the treatment they receive. For example, if we randomly show some people a pro-tax message and others a con-tax message, then before the message, both groups have the same average opinion. So after the message, any difference in opinions must be caused by the message itself. Without random assignment, people who choose to read a pro-tax article might already support taxes, giving a false impression that the article changed minds. Random assignment eliminates that problem. It is the key feature that turns a simple survey question into a causal test. It ensures that the only systematic difference between groups is the experimental manipulation.
12. What is a spillover effect in development evaluation and why does it matter?
A spillover effect happens when a program given to some people also affects others who did not receive it. For example, if a health program treats people for a contagious disease, their neighbors also get healthier. This is good but can mess up causal estimates because the control group indirectly benefits. In an RCT, spillover can make the control group look better than it should, hiding the true effect. To handle this, researchers might design the study so that treatment and control are separated in space, like giving the program to whole villages rather than individuals within a village. They can also measure spillover directly. Ignoring spillover leads to wrong conclusions about a program's impact.