Enduring internal rivalries?

Having recently used the phrase "enduring rivalry" in a recent blog comment on the current crisis in Kenya, I was interested to see that Karl Derouen and Jacob Bercovitch have a paper in the new Journal of Peace Research on "enduring internal rivalries" (gated link here). They import the concept of "enduring rivalry" from the international relations literature and attempt to apply it to civil conflicts. Their coding rules result in a list of 60 enduring internal rivalries.

Enduring rivalries are fruitfully understood as forms of inefficient equilibria. Parties should be cooperating to maximize joint gains, but something prevents them from overcoming costly conflict. These sorts of equilibria may result in large scale violence, but violence may only occasionally punctuate what is otherwise a durable circumstance of conflict. In this way, "enduring rivalries" and long civil wars are not the same thing. One could imagine an enduring rivalry marked by a number of short bouts of violence, for example. To the extent that this is a reasonable characterization of some political circumstances---and I think it is---the challenge is to identify such equilibria when they exist (or have existed).

Enter Derouen and Bercovitch:

[Enduring internal rivalries] denote internal conflicts between a government and an insurgency with at least 10 years of armed conflict in which there are at least 25 deaths – regardless of whether or not these years are consecutive.

They implement this definition by classifying and sometimes lumping together observations from the PRIO armed conflict dataset. They identify 60 enduring rivalries.

Looking at their list, I found the selection of cases to be rough but intuitively reasonable if we are to limit ourselves to situations that have at some point resulted in major violence. But even with this major restriction on cases, the assignment of start and end dates often seemed arbitrary. There also seemed to be some arbitrariness in the decision to define enduring rivalries in terms of factions in some cases and conflicting social groups ("Palestinian insurgents") in others. Perhaps the latter issue is not too important, but it may be indicative of general inconsistencies in the way the original data has been constructed.

More generally, it is dissatisfying to have to use a coding rule that relies on the pre-existence of insurgent groups and a threshold of violence to demarcate cases. This definition would thus exclude years of rivalry that certainly have major socio-economic consequences but precede bouts of violence. In addition, I am not sure what kind of general rule could be used to assign end dates to the more expansive notion of an enduring rivalry that I prefer. If Derouen and Bercovitch are not interested in having us consider such an expansive notion of enduring rivalry, then I am not sure how their effort makes progress over the many civil war duration studies that are already out there.

I was also surprised that the authors chose to study the effects of these enduring rivalries. Bracketing the enormous endogeneity problems here, substantively, do we really need to demonstrate that protracted conflicts are bad or "a distinct class of civil wars"? How about focusing attention on the causes of these enduring rivalries!

Another take on resource dependence and conflict

Christa Brunnschweiler and Erwin Bulte have a new working paper on the effects of natural resource dependence on the risk of civil war onset (link). They begin with the observation that "resource dependence" is likely endogenous to politics and conflict. Exploitation of resources is likely to depend on politics and conflict likelihood. Also, the denominator of the resource dependence variable, gross income, is obviously endogenous. Thus, "simple" regressions of conflict onset on resource dependence will not yield reliable estimates; nonetheless, the literature is replete with these types of regressions. They use IV regression on 5-year panels to overcome this endogeneity problem. Their findings:

Our main findings turn received wisdom upside down. We find that resource dependence is indeed an endogenous variable in conflict regressions, and that properly accounting for this endogeneity removes the statistical association between dependence and conflict. In a follow-up regression we demonstrate, not surprisingly, that a country’s history with respect to war and peace is a significant determinant of resource dependence – clenching our main result. Moreover, we find a significant negative relationship between resource abundance and the onset of war, possibly because of an income effect, suggesting that the label “resource curse” seems misplaced. Resource-rich countries have on average a lower propensity to enter a civil war, but countries that do end up with civil strife (possibly resource-poor ones) will experience increasing dependence on the primary sector.

It's good to see people trying to make further strides to deal with the obvious endogeneity problems in predicting conflict based on economic variables. Let's look at their instruments, though:

The main conditioning variables serving as exogenous instruments for [resource dependence] and ln(gdp) are average openness to trade over the previous 5-year period (openness); a dummy variable for a presidential-type system of government; latitude; percent of land area in the tropics; and distance to the nearest coast or navigable river.

Do any of these plausibly satisfy the exclusion restriction as instruments for resource dependence in predicting conflict onset? I'm not so sure. They would do well to conduct sensitivity analysis along the lines of what was mentioned in this post.

Also, there is some circularity in the way they arrive at their conclusion about the endogeneity of resource dependence to conflict. This conclusion comes from the negative and significant coefficient on "number of peace years" in the IV-first stage regression predicting resource dependence. Why shouldn't we be concerned about the endogeneity of "number of peace years" in this regression?

Anyway, the evidence here is far from a slam dunk, but I actually believe their story: politics and conflict drive the macro economy as much as they are responses to it. But the fact is, this is a really hard claim to demonstrate.

"Indigenous" institutions and colonial origins of development

In a new World Bank working paper, Cambridge historian CA Bayly offers a revision of the "colonial origins of comparative development" story (link to paper). Contrasting Indian and African colonial experiences, Bayly notes that Indian commercial and knowledge institutions were quite sophisticated before the arrival of colonists. Indian elites were thus well-prepared to adapt for their own purposes institutions and practices introduced by the colonists. The British colonists would also employ many locals in administration; eventually Indian administrators would be employed throughout the British empire. As such, the colonial experience did more to boost development capacity in the Indian colonies than in the African colonies. Revising the Acemoglu et al line of reasoning, social conditions as much as environmental conditions that explain the difference in the extent to which institutional transfer succeeded, at least in these cases.

Bayly does not offer a theory of why "indigenous" social conditions were so different in the two regions, but he cites others who discuss ecological factors that favored dense, sedentary agriculture in many parts of India versus more expansive and mobile agriculture throughout Africa. So in a way we are back to ecological factors, but with social consequences in pre-colonial times as intervening factors.

Bayly's contribution is useful in providing clear examples of how the success of "interventions"---in this case, colonial institutional transfer---depend on the recipient social environment. On the one hand, there is some danger here of inspiring people to rush to ad hoc judgments about whether recipient social environments were well-suited to absorb innovations in other cases. On the other hand, it would be similarly foolish to disregard the potential constraints associated with low "absorption capacity" of recipient social environments. Grist for the mill...

Evidence of "Perverse Punishment" and "Collectivist Socialization" Effects

Simon Gachter and Benedikt Herrmann present results in a new working paper on their public goods games experiments in urban and rural Russia (link to paper):

Abstract

We report evidence from public goods experiments with and without punishment which we conducted in Russia with 566 urban and rural participants of young and mature age cohorts. Russia is interesting for studying voluntary cooperation because of its long history of collectivism, and a huge urban-rural gap. In contrast to previous experiments we find no cooperation-enhancing effect of punishment. An important reason is that there is substantial punishment of high contributors in all four subject pools. Thus, punishment can also undermine the scope for self-governance in the sense of high levels of voluntary cooperation that are sustained by sanctioning free riders only.

Elinor Ostrom's and Ernst Fehr's experiments in the 1990s caused quite a stir in the social sciences because they showed that people are willing to pay to punish others for not contributing to public goods. This type of pro-social behavior defied the predictions of "Nash"-type rational behavior in public goods games.

A line of skeptical inquiry since then has been to look into whether the results of these experiments are themselves endogenous to the social environment. Maybe Fehr's subjects---students from a Swiss university---were not products of a social environment that would make them representative of humankind, for example. Gachter and Herrmann's paper provides more evidence that this is indeed the case. For Russia, Gachter and Herrmann propose that varying exposure of "collectivization" programs would leave varying marks in the behavior of subjects. They find some supporting evidence: older people and rural people, both of whom were exposed to higher "doses" of collectivization, tend to contribute at higher rates in the public goods games.

The experiments thus become measuring devices for the effects of aspects of the social environment. With this being the case, researchers are now faced with an additional challenge. Not only do they have to work out the challenges of designing and implementing the experiment, but they have to do so in places where the effects of particular aspects of the social environment are identifiable. Behavioral experiments themselves do not produce generalizable results if the effects measured in the experiments are conditioned by unidentifiable environmental variables. While Gachter and Herrman's results are intriguing, should we accept that age and rurality map uniquely to collectivization experiences? Probably not. The lesson for me: when we are interested in developing generalizable knowledge about human behavior, behavioral experiments are strong measuring devices, but are nonetheless subject to identification problems associated with observational studies.

Kenya elections and violence, III

An article from the IRIN news service (linked here) suggests that the following factors were central in motivating people to participate in the violence:

(1) Frustration associated with high socio-economic inequality throughout the country, as documented in a report by Kenya's Society for International Development (link here), using UNDP socio-economic data.

(2) Perceptions of unfair distribution of opportunities. The article claims,

Ethnicity came into play during the election violence because of the widespread perception that those who fared best under Kibaki were his own Kikuyu group, the country’s largest, which dominated politics and the economy both under his administration and that of founding president Jomo Kenyatta.

(3) Frustrated ambition of the current young adult generation, which has received more education than previous generations:

Kenya’s youth in particular, who make up a majority of the population - and of those who rioted - feel the most let down. Improved education gave them hope of a better life than their parents’, hope that was dashed, according to Kwamchetsi Makokha of Nairobi-based communications consultancy Form and Content.

“Under colonialism, it was almost a slave labour system which grew up in the early days of the coffee estates. After independence [in 1963], the white master was simply replaced by the black master. A lot of young people who got a bit of education could not see themselves working for pittances as farm labourers. They started drifting to the cities where the opportunities are not enough to accommodate all of them. You have this massive influx of people who just can’t find work,” he told IRIN.

Nor can they find a political voice, he added. “The common Kenyan citizen who does not have money or property does not have a say in how Kenya is organised. They never have. It’s always been about what car you drive, where you live, and then you have more rights than other people.”

"Huntington 1968" should be ringing in peoples minds as they read that.

(4) Frustration with the corrupt practices of the Kibaki regime: "Another ingredient in this combustible mix is corruption, which Kibaki pledged to eradicate but which under his rule, according to analyst and author Gerard Prunier, 'reached new heights, matching some of the excesses of the Moi years'. "

One thing that strikes me in reading this is the extent to which grievances are perceptions. That's where the challenge comes in conducting rigorous analysis of the link between inequality, favoritism/discrimination, and political upheaval. Perceptions may cause participation in an uprising, but the perceptions themselves are caused by strategies of "political entrepreneurs" and objective conditions. I sense that a lot of debate among analysts and researchers over "true" motivations centers on disentangling the relative contribution of these factors in causing perceptions injustice. Of course, there is another view prevalent these days that much of the participation in upheaval is somehow "opportunistic", but to me, those interpretations have to first explain where from the "opportunity" for an uprising comes. I think any serious thought on that question would lead the analyst back to considering perceptions of injustice.

Kenya elections and violence, II

Discussion of events in Kenya on Ryan Sheeley's blog continue. Ryan discussed the idea that the restoration of order may need to flow from the local to the national. I posted a comment that raises some questions about this view:

I put a comment on Chris's blog yesterday that may challenge the faith put into "local solutions." First, if participation in violence is largely determined by the extent to which one is frustrated by discrimination, favoritism, or other barriers to realizing one's potential, then "local solutions" are only available to the extent that the relevant barriers operate at the local level. But is this the case here? Or are we talking about macro level barriers (e.g. mass discrimination organized across ethnic lines)?

Second, in my own examination of conflict histories in developing societies, I have found that "local solutions" tend to be conservative. That is, local institutions tend to police the barriers against which mass protest is rallying. Without being too teleological, one could say that their erosion is part of the modernization process. Thus, CDF mobilization in Sierra Leone, for example, was a conservative, counter-insurgent response to the RUF. The "local solution" was effective because this conservative movement was both allied with foreign interveners and sufficiently effective in mobilizing people to preserve the status quo. As a contrasting example, rebel mobilization in Burundi thoroughly transgressed "local institutions" precisely because these local institutions were erected to police, at the local level, the type of mass discrimination that rebels sought to overthrow. No "local solution" was imaginable in this case. Thus, whose interests would "local solutions" in Kenya satisfy?

UPDATE: See the comments linked below for a very informative response from Ryan.

Kenya elections and violence

On their blogs, Chris Blattman and Ryan Sheeley have been offering updates and links to other blogs on the Kenyan electoral crisis. I posted the following comment on Chris's blog:

I wonder about the deeper background to the current crisis.

One gets the sense that an all too common story might be at work here:

Undiversified, aid-dependent economy means that control of the state equals control over significant assets and opportunities. Access is conditional on relationship to incumbent, who is thus custodian of exclusionary political economy. For some reason, incumbent loses control over electoral dynamics, which presents a secular opportunity for the excluded to seize control of assets and opportunities. Recognizing what is at stake, incumbent tries to prevent control from being pried from his grasp. Fighting ensues.

Something like this story is what we heard about Burundi in 1993 during the course of our research there over the past 2 years. It's also like the story one reads about Rwanda in 1959 or even Congo-Brazzaville since 1997. It echoes recent stories told about Bolivia, heck even Venezuela for some... It bears resemblance to the "revolutionary politics" story that has been formalized by Carles Boix, Daron Acemoglu and James Robinson, and Thad Dunning, among others.

I wonder whether this "exclusion" or "mass discrimination" lens is relevant here. If one were to peer into the records in the education system, for example, would one find overrepresentation of one group or another? If so, there are implications for how external aid should be used as leverage for dealing with the type of exclusion that may be at the heart of the crisis.

I'd be interested in hearing responses to the applicability of this lens.

Survey Experiments: Dictator Game in a Mass Survey

Another clever way of embedding a behavioral experiment into a survey:
Rene Bekkers. 2007. "Measuring Altruistic Behavior in Surveys: The All-or-Nothing Dictator Game." Survey Research Methods, Vol 1, No 3 (open access link here). In surveys, respondents are often given some kind of compensation for their participation. This survey did so as well, but Bekkers also gave respondents the opportunity to donate their payment to a charity rather than keep it for themselves. Here is the abstract:

A field study of altruistic behaviour is presented using a modification of the dictator game in a large random sample survey in the Netherlands (n=1,964). In line with laboratory experiments, only 5.7% donated money. In line with other survey research on giving, generosity increased with age, education, income, trust, and prosocial value orientation.

Bekkers's method here also differs from common dictator game approaches in that the subjects were allocating "earned" endowments rather than windfall endowments. Past research suggests that such determinants of "asset legitimacy" affect what subjects do with their endowments. Note again that it is the survey process itself that allows for this approach to be taken. Of course, since there is no variation in this treatment, we are limited in how much we can learn about such phenomena. Maybe that was a missed opportunity?

In a complementary analysis of self-reported contributions to charities, Bekker follows Smith, Kehoe, and Cremer (1995) ("The private provision of public goods") in using 2SLS to estimate correlates of giving conditional on having given at all. But he claims to have assigned some kind of randomized treatment in the survey that was used to satisfy the exclusion restriction for the first stage of 2SLS. There seemed to be a bit of hand waving on this, and results from the first stage are not included in Table 2. I found this presentation of 2SLS results to be too vague.

Costs of Conflict

An important issue in the study of violent conflict is the nature of the economic costs that are imposed. On the one hand, instability drives away investment, and resources committed to fighting and the destruction wrought by conflict both steal from productive capacity. On the other hand, it is conceivable that political change ushered in via conflict can result in redistribution of assets and opportunities that increase long run efficiency. The economic costs of conflict are conventionally understood as fundamental in determining whether protagonists decide to fight.

A major difficulty in addressing this issue is measurement. A new paper by Zussman, Zussman, and Oregaard in the current Economica (gated link here) shows how asset market data can be used to measure the economic effects of conflict. Their key methodological contribution is a way of identifying "turning points" in financial time-series. They apply this method to data from Israeli and Palestinian asset market series. They find that their methodology does a good job at identifying turning points in the market series that correspond to key events in the Israeli-Palestinian conflict. Aggregates over periods between turning points can be used as summaries of economic costs of conflict. Doing so, they find that "rough calculation based on the results of our analysis yields a drop of 22% in the value of the Tel Aviv Stock Exchange resulting from the outbreak of the Intifada and an increase of 25% in market value arising from the adoption of the Road Map peace plan."

So can we interpret this as saying that about a quarter of Israeli asset market activity was sensitive to the conflict? If so, one wonders what factors contribute to the extent to which markets are sensitive to conflict. In addition, to what extent does this number characterize general economic sensitivity to conflict in Israel? To the extent that markets are made more resistant to conflict, are incentives to fight altered?

Volunteering, Happiness, and a Clever Natural Experimental Design

Here's the abstract from an interesting paper by Stephen Meier and Alois Stutzer in the current issue of Economica (gated link here)

Stephen Meier and Alois Stutzer. "Is Volunteering Rewarding in Itself?" Economica 75(297):39–59.

Abstract

Volunteering constitutes one of the most important pro-social activities. Following Aristotle, helping others is the way to higher individual wellbeing. This view contrasts with the selfish utility maximizer, who avoids helping others. The two rival views are studied empirically. We find robust evidence that volunteers are more satisfied with their life than non-volunteers. The issue of causality is studied from the basis of the collapse of East Germany and its infrastructure of volunteering. People who lost their opportunities for volunteering are compared with people who experienced no change in their volunteer status.

That's a clever research design.

Error Correction Models for Elections

Virginia and I were looking at how to implement error correction models (ECMs) to study the stability of vote shares for parties and incumbents. Thought I would share some of the material in case others are working on these topics too. The main paper that we looked at was the McDonald and Best (2006) Political Analysis paper (gated link here). They use ECMs to study variation in the stability of incumbent vote shares under different electoral systems. It's an interesting approach, although Virginia and I agreed that they should have used seemingly-unrelated regressions to estimate the trends for the parties in the different countries (you have to look at the paper for that to be meaningful).

Also, here's a link to a little tutorial that I wrote up that goes through an ECM example in Wooldridge Introductory Econometrics (a.k.a. "baby Wooldridge"). It shows full implementation of the Engle-Granger two-step method as well as the Banerjee et al one-step method. The example in the tutorial is a standard type of ECM for two conitegrated series. It is a bit more sophisticated than what is going on in the McDonald and Best paper, which only evaluates error correction in a single series.

Does "war in mixed strategies" make sense?

At yesterday's political economy workshop, our own Massimo Morelli presented a working paper that he co-authored with Matthew Jackson entitled "Strategic Militarization, Deterrence, and Wars" (paper is here). The centerpiece of the paper is an analysis of a mixed strategy equilibrium involving states playing "hawkish," "dovish", and "deterrence" strategies, with wars occurring when "hawkish" and "dovish" strategies interact.

This is the second paper that we've seen at the political economy workshop this Fall where fighting has occurred only in mixed strategies. The other paper was a working paper by Ernesto Dal Bo and Robert Powell on civil conflict resulting from a situation where governments hold private information on the size of a centrally controlled endowment from which contenders seek a share (paper is here).

This raises all the questions about whether such mixed strategy "fighting" equilibria are plausible characterizations of behavior. In the Dal Bo and Powell, paper, contenders mix over fighting and not fighting to "keep the government honest". However, the contender's decision to fight comes after she has already received her share of the endowment from the government. For the equilibrium to be believable, we have to accept that even after the contender's share has been paid out, she would still with positive probability initiate an insurgency that offers no additional gain. This strikes me as implausible.

In the Morelli and Jackson paper, things are a little better insofar as the timing of the game does not conflict so much with the logic behind the mixed strategy. But it's still a case where wars are initiated as the result of joint randomization of strategies.

However, Massimo argued for another interpretation. Rather than thinking in terms of randomizing strategies in the context of a bilateral interaction, think of one central actor facing many adversaries. Then, we can imagine a mixed strategy as being one where the central actor is playing a different strategy with each adversary such that the distribution of strategies conforms with the equilibrium "mix". My thought was that this interpretation was stretching things a bit. This interpretation would seem to require that the adversaries all act as if they are playing a bilateral game with the central actor; but clearly in any real world situation, the adversaries would condition their behavior on what the central actor is doing vis-a-vis the other adversaries. So, to be convinced, I would have see that the same equilibrium "mix" holds in a compelling, respecified game with one central actor and multiple adversaries and with the central actor choosing war with a subset of the adversaries in pure strategies. I just don't think that something like war makes any sense as an element of a mixed strategy.

Caillaud and Tirole on Group Persuasion

A paper by Caillaud and Tirole in the current AER (gated link) extends the analysis of sender-receiver games to study group persuasion. They motivate the paper with the example of the sponsor of a policy proposal attempting to persuade a committee. Here's the abstract:

The paper explores strategies that the sponsor of a proposal may employ to convince a qualified majority of members in a group to approve the proposal. Adopting a mechanism design approach to communication, it emphasizes the need to distill information selectively to key group members and to engineer persuasion cascades in which members who are brought on board sway the opinion of others. The paper shows that higher congruence among group members benefits the sponsor. The extent of congruence between the group and the sponsor, and the size and the governance of the group, are also shown to condition the sponsor’s ability to get his project approved.

And here are some of their counterintuitive results:

We showed that adding veto powers may actually help the sponsor, while an increase in external congruence [of preferences among the sponsor and committee members] may hurt him; that a [committee member with more congruent preferences to the sponsor] may be worse off than an a priori more dissonant member; and that, provided that he can control channels of communication, the sponsor may gain from creating ambiguity as to whether other members really are on board. Finally, an increase in internal congruence [of preferences among committee members] always benefits the sponsor.

On a technical note, they do not specify a game form but rather study the communication mechanisms that are optimal for the sponsor in his effort to win approval from the committee. The empirical content of the model includes some comparative statics on how size, preference congruence, and voting rules affect the likelihood of proposal acceptance and thus the stability of the policy status quo.

Non-reponse and False Response in Corruption Surveys of Firms

Abstract from a new working paper by Rahman, Li, and Jensen:

Heard melodies are sweet, but those unheard are sweeter : understanding corruption using cross-national firm-level surveys

2007-11-01

By: Rahman, Aminur; Li, Quan; Jensen, Nathan M.

http://d.repec.org/n?u=RePEc:wbk:wbrwps:4413&r=dev

Since the early 1990s, a large number of studies have been undertaken to understand the causes and consequences of corruption. Many of these studies have employed firm-level survey data from various countries. While insightful, these analyses based on firm-level surveys have largely ignored two important potential problems: nonresponse and false response by the firms. Treating firms ' responses on a sensitive issue like corruption at their face value could produce incorrect inferences and erroneous policy recommendations. We argue that the data generation of nonresponse and false response is a function of the political environment in which the firms operate. In a politically repressive environment, firms use nonresponse and false response as self-protection mechanisms. Corruption is understated as a result. We test our arguments using the World Bank enterprise survey data of more than 44,000 firms in 72 countries ! for the period 2000-2005 and find that firms in countries with less press freedom are more likely to provide nonresponse or false response on the issue of corruption. Therefore, ignoring this systematic bias in firms ' responses could result in underestimation of the severity of corruption in politically repressive countries. More important, this bias is a rich and underutilized source of information on the political constraints faced by the firms. Nonresponse and false response, like unheard melodies, could be more informative than the heard melodies in the available truthful responses in firm surveys.

This is an important type of analysis. By construction, the survey serves as an experiment to test a behavioral model of firms' willingness to report corruption. This type of "incidental survey experiment" is a nice way to perform secondary data analysis. Not only do we learn something about behavior, but we also learn something about the data itself.

When is something a "0" versus an "NA"?

In a recent talk in the department, a presenter wanted to code leadership transition outcomes according to a binary rule, where "transition with punishment" outcomes were coded as 1's and "transition with no punishment" outcomes were coded as 0's. The "transition with no punishment" cases included natural death or assassination while in office.

It seemed to me that this coding didn't make sense. When you have death in office, the case should be considered as censored, no? In these cases, you are not able to observe whether the outcome is truly "punishment" or not. Think about it. So the "death in office cases" are not really cases of "no punishment" but rather "not observed." Then, one needs to decide whether these cases should just be dropped from the analysis or whether a selection correction should be included.

I claim that a terrible thing to do is to lump censored observations into the "0" category. Here's my reasoning. We know that simply dropping cases where there is censored data (listwise deletion) leads to bias when the likelihood of censoring is dependent on the outcome (Y). But when censoring likelihood is dependent only on the explanatory factors (the X's) of interest, listwise deletion is not biased. But labeling censored cases as 0's can lead to bias if censoring is associated with either Y or the X's.

To see an example of how this works, consider the following contingency table:

Table 1: Table to Estimate Pr(Y=1|X=0) and Pr(Y=1|X=1) When No Censoring Is Present

----+---+---+
Y=1 | 2 | 2 |
----+---+---+
Y=0 | 2 | 2 |
----+---+---+
----|X=0|X=1|
----+---+---+

The relationship shown in Table 1, which is the "true" relationship since all data is observed, is that Pr(Y=1|X=0) equals Pr(Y=1|X=1). Now suppose that there is a 1/2 chance that data will be missing when X=0 and a 0 chance that it will be missing otherwise. Listwise deletion would produce the following table:

Table 2: Table to Estimate Pr(Y=1|X=0) and Pr(Y=1|X=1) Given Censoring Dependent on X and Listwise Deletion

----+---+---+
Y=1 | 1 | 2 |
----+---+---+
Y=0 | 1 | 2 |
----+---+---+
----|X=0|X=1|
----+---+---+

Table 2 leads us to infer correctly that Pr(Y=1|X=0) equals Pr(Y=1|X=1). However, if we assume the same censoring mechanism, then labeling censored observations as "Y=0" gives the following:

Table 3: Table to Estimate Pr(Y=1|X=0) and Pr(Y=1|X=1) Given Censoring Dependent on X and Labeling Censored Obs as Y=0

----+---+---+
Y=1 | 1 | 2 |
----+---+---+
Y=0 | 3 | 2 |
----+---+---+
----|X=0|X=1|
----+---+---+

Table 3 leads us to infer incorrectly that Pr(Y=1|X=0) does not equal Pr(Y=1|X=1). If the "true" relationship is one in which Pr(Y=1|X=0) does not equal Pr(Y=1|X=1), things can be similarly messed up (try it with simple contingency table examples). What a mess...

Presentation on "Globalization's Losers"

In the department today, Yotam Margalit presented some results from his study on "Globalization's Losers: Trade, Culture and the Politics of Discontent." (Abstract of his research is here.) He was looking at the chain linking socio-economic attributes, negative attitudes toward globalization and integrationist policies, and then voting for right or left parties. The data that he presented suggested that predictions from the ol' Heckscher-Ohlin model do seem to conform to reality if we relate the model to income levels, assuming that owners of relatively scarce factors (conditional on economic openness) correspond to the poor in developed countries and the middle-class or rich in developing countries. But the predictions do not fare well when we relate the model to the ideological leanings of voters and parties, assuming that protectionism should be the cause of the left in developed countries and the right in developing countries. The latter prediction---specifically, that antiglobalization forces should be decidely on the left in developed countries---is not borne out by the data. Why? Margalit claims that there is a second dimension, one that he calls "perception of cultural threat," which intervenes. Antiglobalization forces, it seems, are animated by leftist economic concerns but rightist "cultural concerns." When you put the two together, you get antiglobalization voters and parties that are all over the single-dimensional ideological spectrum. When you permit parties and voters to locate themselves freely in a two-dimensional ideological space, something that proportional representation comes close to allowing, you get clustering in the "cultural right, economic left" region of the two-dimensional space.

Margalit's analysis leads us to conclude that what he labels as "cultural concerns" disrupt what otherwise would be a clean mapping between economic interests and political expression of dislike for globalization. But we are left to ponder, what exactly are these "cultural concerns"? Are they really just idiosyncratic, country-specific factors---some kind of error term---that must be studied on a case by case basis? Such was the way that Margalit responded to questions about what the cultural factors were. Does that make sense? Or are there systematic forces---racism? religion? generic fear of change?---at work here?

"Cellphones Challenge Poll Sampling" (NYT)

Interesting article on how survey sampling in the US is being affected by the fact that many people no longer keep land-line telephones and rather have only cell phones. The article states that "the issue came up [during polling for the US elections] in 2004, but cellphone-only households in 2003 were 3 percent of the total. They now run 16 percent, according to Mediamark Research." More from the article:

According to data from the Centers for Disease Control and Prevention’s National Health Interview Survey, adults with cellphones and no land lines are more likely to be young — half of exclusively wireless users are younger than 30 — male, Hispanic, living in poverty, renting a residence and living in metropolitan regions.

The Pew Research Center conducted four studies last year on the differences between cellphone and land line respondents. The studies said the differences were not significant enough to influence surveys properly weighted to census data. With the increase in cellphone-only households, that may not be the case next year. Researchers, including the New York Times/CBS News poll will test that by incorporating cellphones in samples.

The estimates in the Health Interview Study suggest that cellphone-only households are steadily increasing.

“If the percentage of adults living in cell-only households continues to grow at the rate it has been growing for the past four years, I have projected that it will exceed 25 percent by the end of 2008,” Stephen J. Blumberg, a senior scientist at the National Center for Health Statistics, wrote in an e-mail message.

The American Association for Public Opinion Research has been examining the question and formed a group to study it. The association says it will issue its report early next year.

Clearly the limited impact for election polling does not apply to surveys that attempt to measure other population parameters. The weighting fix proposed by Pew is adequate when estimating simple population parameters (e.g. proportions), but things get much more dicey when we move to a regression framework. (See this article by Andrew Gelman for a recent take.)

Climate Change Polls Summary

WorldPublicOpinion.org has a summary of recent polls on attitudes around the world toward climate change:

A new analysis by WorldPublicOpinion.org of 11 recent international polls conducted around the world shows widespread and growing concern about climate change. Large majorities believe that human activity causes climate change and favor policies designed to reduce emissions.

In most countries, majorities see an urgent need for significant action. For example, a recent poll for the BBC by GlobeScan and the Program for International Policy Attitudes (PIPA) found that majorities in 15 out of 21 countries felt that it was necessary to take “major steps, starting very soon” to address climate change. In the other six countries polled, opinion was divided over whether “major” or “modest steps” were needed. Only small minorities thought no steps were necessary.

The analysis included polls from the BBC/GlobeScan/PIPA, the Pew Research Center, GlobeScan, WorldPublicOpinion.org/Chicago Council on Global Affairs, the German Marshall Fund, and Eurobarometer. (Link to report.)

Seems like this data would provide a nice starting point for a global public goods provision analysis. Does it make sense to study how regime structure mediates the way public interest is channeled into action? How would one structure the analysis? Or does it only make sense to study dynamics associated with responses to climate change in terms of global-level bargaining? Or maybe a two-level game framework would make sense---could we use these data, interacted with domestic regime, to estimate the size of the domestic "win-set"?...

A Take on the "Plausible Instrumental Variables" Debate

In our "quantitative methods in poli sci" seminar today, Andy Gelman and Piero Stanig debated the importance of random assignment of instrumental variables for valid causal inference. The claim being debated was whether it is true that random assignment (literally random---i.e. picking balls from urns, coin flips, etc.) of the instrument in addition to the exclusion restriction on instrumental variables (i.e. no direct effects of the instrument on outcome) and significant first stage are all required to draw valid causal inference from IV regressions. Here's my take on the discussions. If any of you had other interpretations, please comment!

First, it was plain for everyone to see that random assignment of the instrument is not sufficient to ensure that the exclusion restriction is satisfied and therefore is not sufficient to provide leverage for causal inference. As an example, Vietnam draft lottery numbers used as instruments for military service by Angrist, Imbens and Rubin (1993) (gated) were randomly assigned, but the exclusion restriction may have been violated if there were other consequential effects resulting from one's draft lottery number other than military service (e.g. other lifestyle changes that may have resulted from receiving a particular lottery number). Also, it hardly needs to be said that random assignment is not sufficient to ensure that the "significant first stage" assumption holds.

Second, it was understood that random assignment helps to ensure that values of the instrument are ignorable relative to values of the outcome---i.e., values of a randomly assigned instrument are surely not determined by potential outcomes---i.e. values of the instrument are exogneous relative to the outcome (three ways of saying the same thing).

Third, random assignment helps to ensure that the variation in the values of the instrument are not the result of some unobserved factor determining the values of the instrument, the explanatory variable being instrumented, and the outcome. It is in this broader web of relations that random assignment helps to ensure that the exclusion restriction is, at least in part, satisfied. This was probably the most important point that came up during the discussion. This is associated with the "it's culture" argument that is often used to challenge results in comparative politics, with culture being an unobserved factor that simply determines everything. The idea here is that without random assignment, one needs to think hard about the mechanisms through which changes are brought about in the value of the instrumental variable. Is it reasonable to believe that those mechanisms are not sneaking exclusion restriction violations in through the back door? Also, are we confident that a change in the instrument as a result of that mechanism will then produce a change in the explanatory variable in the manner resembling what was estimated in the first stage?

But neither the second or third points above would have us conclude that assignment by way of genuine randomization (picking balls from urns, coin flips, unpredictable weather patterns, etc.) is necessary for the "ignorability", "exclusion restriction", and "significant first stage" assumptions to hold. But nonrandom assignment means that you will have to think harder about whether these requirements are met. I think in the end everyone was willing to accept that.

Aside from the debate, Piero presented an interesting application of some sensitivity analysis methods to test for the consequences of "slight" violations of the exclusion restriction. He pointed us to this working paper by Conley, Hansen and Rossi for more details.

Also, Bob Erikson and Andy proposed using some "common sense" in thinking through whether your IV methods are valid: Suppose you are using rainfall as an instrument for economic growth as a predictor of civil war. Say to yourself, "I just showed that rainfall is associated with civil war." Now think a bit, are you led then to say, "Ah yes, it must be because rainfall determines growth, which we have reason to believe is related to civil war"? Do we believe that "must"?

UPDATE: Andy Gelman has his own thoughts on the debate here.

Applied Bargaining Models and Mechanism Design

Abstract from an interesting working paper, "Uncertainty and Incentives in Crisis Bargaining: Game-Free Analysis of International Conflict, "by Mark Fey and Kristopher Ramsay (available here):

The formal literature on international conflict has identified the combination of uncertainty and the incentive to misrepresent private information as a central cause of war. But there is a fundamental problem with using game-theoretic models to formulate general claims such as these---whether and to what extent a result that holds in a particular choice of game form continues to hold when different modeling choices are made is typically unknown. To address this concern, we present techniques from Bayesian mechanism design that allow us to establish general "game-free" results that must hold in any equilibrium of any game form in a broad class of crisis bargaining games. We focus on three different varieties of uncertainty that countries can face and establish general results about the relationship between these sources of uncertainty and the possibility of peaceful resolution of conflict. We find that in the most general setting of uncertainty about the value of war, there is no equilibrium of any possible crisis bargaining game form that allows the unilateral use of force that completely avoids the chance of costly war.

This is part of a series of papers that Fey and Ramsay have been writing on so-called "game free" analysis of crisis bargaining. The approach is appealing because it lends itself to consideration of classes of conflict resolution mechanisms that ought to be robust to changes in the bargaining procedure. As Fey and Ramsay describe, this type of analysis is especially relevant in crisis bargaining contexts. It is often the fluidity of the bargaining context that makes crisis bargaining situations distinct from institutionalized bargaining in, say, legislatures. (Although, one may argue that distributions of power may lend some structure to the bargaining situation. For example, the dynamic "ultimatum game" structure of Acemoglu and Robinson's transitions game, Boix's transitions game, and Fearon's civil war settlement game derive from an assumption about which actors have the capacity to make proposals.)

Interestingly, I came to this article after having assisted Macartan in preparing a draft of a review article on coalitional analysis. That article gives considerable attention to cooperative game theory and the Nash program. I see some similarity between what Fey and Ramsay are doing and the Nash program---basically, defining classes of games that implement axiomatic solutions.