Experimental Evidence on Transparency Interventions

Although well executed and informative, the observational studies cited both in favor of transparency and against it have limitations that inhibit causal inference and the ability to extrapolate actionable policy lessons. Simultaneity bias haunts direct causal identification. For instance, it is possible that voters who care deeply about better public goods delivery make more effort to inform themselves about politician behavior. Unobserved heterogeneity is also a threat, as the same features that make a state more open to diverse media outlets can also lead to better selection of politicians. Of course, these problems are inherent in most work, and many of these articles are quite careful about causal identification. More importantly, from a policy standpoint, the observational work cited previously offers little insight into what can be done to improve political transparency going forward. Both media penetration and informed voters are the result of long-term, multifaceted historical processes, but what opportunities are available to countries and localities not blessed with such endowments?

To this end, a new literature has begun to use experimental approaches to better sort out the relationship between policy changes that increase transparency and political performance. Carey (Reference Carey2010) shows that groups of students placed under laboratory conditions responded to openness with more generosity to their peers. Humphreys and Weinstein (Reference Humphreys and Weinstein2007) take a slightly different tack by randomizing which Ugandan parliamentarians were treated with a scorecard of information recording their attendance, participation, legislative initiatives, and contact with their constituencies. In a pilot analysis, the authors found that treated delegates were 50% more likely to participate in subsequent parliamentary sessions. Evidence has also been identified for the selection hypothesis. Voters supplied with criminal histories (Banarjee et al. Reference Banarjee, Green, Green and Pande2010), district budget allocation data (Paler Reference Paler2011), and information about incumbent corruption (Chong et al. Reference Chong, De La O, Karlan and Wantchekon2010), responded by punishing and rewarding politicians accurately based on the new information.

Exporting Transparency to Authoritarian Regimes

The growing number of experimental studies on transparency are encouraging, but all take place in the context of an electoral democracy. Even Uganda, which is coded as only “partially free” on the most recent Freedom House Index, has held multiple, competitive elections. How might such experiments fare in nondemocratic or competitive authoritarian regimes, where the relationship between citizens and voters is less understood? This is becoming an increasingly relevant question as more and more authoritarian regimes adopt transparency initiatives at the behest of foreign donors. The strength of the theoretical arguments for increased transparency has impacted a number a number of international organizations and NGOs, which seek to improve public service delivery in developing country settings by funding interventions that enhance the transparency of policymaking, particularly in national legislatures. In authoritarian settings, practitioners have speculated that transparency may achieve these positive benefits even in the absence of a functioning electoral democracy. Though their number is difficult to measure precisely, McGee and Gaventa (Reference McGee and Gaventa2010), in a review of such projects for the Institute of Development Studies, estimated the number of transparency initiatives to be in the thousands (3), noting that such interventions have increased in the past decade, “as key ways to address both developmental failures and democratic deficits (1).”

Better information about political behavior may allow media and civil society to report on political abuses and embarrass political leaders, creating “millions of auditors” (Kaufman 2004, 21), as a substitute for direct accountability through elections and voters (Peruzzotti and Smulovitz Reference Peruzzotti and Cataline2006; Smulovitz and Perruzzotti Reference Smulovitz and Peruzzotti2000). This logic builds off the fire alarm style of public monitoring first emphasized by McCubbins and Schwartz (Reference McCubbins and Schwartz1984). In complex economic and social systems, it can be costly and inefficient for central government officials to perform police-patrol-style monitoring, in which agents personally inspect every action made by subordinate government officials and delegates. An alternative strategy is to move to a fire alarm approach, where citizens or media “pull the alarm” when they see wrongdoing. This allows the central government to respond to the abuse without the cost of daily inspection of every subordinate government activity. For fire alarm monitoring to be successful, however, states must increase the openness of information to citizens so they can play this monitoring role. However, this approach relies on the untested assumption that alternative mechanisms can substitute for downward accountability to citizens through elections (Joshi Reference Joshi, McGee and GaventaSussex2010). Two variants of this assumption are employed by practitioners. A weak version suggests that public shaming creates a sufficient incentive for the subordinate actors (delegates) to change their behavior. A stronger version of the assumption is that electoral accountability can be replaced by upward accountability to elite government officials, assuming that, once alerted, central officials will punish self-dealing and corruption. This same logic lies behind the Chinese government's decision to enact its own Open Governance Initiative (OGI) with assistance from the Yale Law Center (Horsley Reference Horsley2008; Ma and Wu Reference Ma and Wu2011). Although the OGI was initiated by the Chinese state itself and applies to subordinate officials, international donors have also sought to export the logic of this approach to legislatures, arguing that the public shaming of parliamentary delegates or the threat of central punishment will incentivize delegates to change their behavior (Joshi Reference Joshi, McGee and GaventaSussex2010). In short, there is a belief that the incentive effect of transparency can still apply even in polities that have nonfunctioning or highly flawed electoral systems (Bauhr and Grimes Reference Bauhr and Grimes2011, 1).Footnote ¹

However, based on the arguments of transparency skeptics as well as research on authoritarian legislatures, there is reason to believe that the logic linking transparency to parliamentary performance does not travel well to authoritarian settings, even in quasi-democratic institutions, such as hegemonic regimes with universally elected national legislatures (Lindstedt and Naurin Reference Lindstedt and Naurin2010). The critical difference is that nominally democratic institutions may serve a starkly different role in authoritarian countries (Gandhi and Lust-Okar Reference Gandhi and Lust-Okar2009), even though they carry the same names. It is a well-known fact that the vast majority of authoritarian regimes have legislatures and elections. Hegemonic party and single-party regimes accounted for more than 60% of the nondemocracies in 2006 (Magaloni and Kricheli Reference Magaloni and Kricheli2010), and virtually all those countries had national legislatures during the post-Cold War period. Monarchies and personalist regimes are also likely to have parliaments (Wright Reference Wright2008). Elections are similarly widespread. Between 2000 and 2008, of the 172 existing independent states with a population greater than 500,000, only six countries failed to hold some form of direct national election (Hyde and Marinov Reference Hyde and Marinov2012).Not only is the proliferation of elections and parliaments widespread, but also empirical analyses have consistently shown that authoritarian regimes with elections perform better on a range of critical outcome variables than authoritarian states that have yet to adopt them. Contributors to this literature have demonstrated a strong association between having a national legislature and regime longevity, reduced violence, and economic growth (Gandhi Reference Gandhi2009; Gandhi and Przeworski Reference Gandhi and Przeworski2007; Gandhi and Vreeland Reference Gandhi and Vreeland2004; Geddes Reference Geddes2006; Wright Reference Wright2008). The benefits of nominally democratic institutions in authoritarian regimes have rarely been attributed to increased delegate–voter linkage. Authors in the authoritarian institutions literature argue that the use of elections and assemblies has a separate set of causal logics (Gandhi and Lust-Okar Reference Gandhi and Lust-Okar2009) that involve either signaling or co-optation of a potential opposition.

Signaling theories suggest that elections can deliver supermajorities for regime-backed candidates, thus bolstering their own legitimacy and preventing the opposition from mounting a challenge (Geddes Reference Geddes2006; Magaloni Reference Magaloni2007; Simpser Reference Simpser2005). Co-optation theorists, in contrast, argue that authoritarians relying too heavily on repression for survival become dependent on their security apparatus (e.g., police force or military), which carries out the state's heavy-handed enforcement. Maintaining authority in this manner places a large share of resources in the hands of security elites, who could turn against the regime leaders down the road (Gandhi Reference Gandhi2009; Haber Reference Haber, Weingast and Wittman2006). Co-optation theorists therefore argue that a less dangerous long-term strategy is to allow groups from outside the inner circle to have a formal say in the policymaking process through the use of quasi-democratic institutions. Elections allow leaders to identify the most popular local notables or potential opposition forces (Boix and Svolik Reference Boix and Svolik2007). Once they are identified, the dictator can placate these elites by giving them some say over policymaking (Gandhi Reference Gandhi2009; Gandhi and Przeworski Reference Gandhi and Przeworski2006; Reference Gandhi and Przeworski2007) and/or access to rents (Lust-Okar Reference Lust-Okar2006) through membership in a national parliament. For co-optation theorists, formal legislatures are more effective than simply negotiating directly with outside groups, because the co-optation takes place in a more stable, institutionalized environment than would be the case in an informal, ad hoc arrangement. In terms of the role of the co-opted legislators, although they may be highly active, the primary role of national parliaments is to serve as forums where “demands can be revealed without appearing as acts of resistance . . . and where the resulting agreements can be dressed in a legalistic form and publicized as such” (Gandhi and Przeworski Reference Gandhi and Przeworski2007, 1282). Thus, a critical feature of the co-optation theory is that disagreements by delegates and important sectors of society can be presented in a controlled and unthreatening manner that will not generate larger protests.

If the positive benefits associated with legislatures are due to this co-optive exchange, transparency could potentially undermine the delicate balance worked out between leaders and potential opposition. More transparency could allow intense and volatile debates to spill outside of the contained forum and into environments where it is more difficult to identify solutions through side deals and particularistic payments. Delegates may find it more difficult to criticize regime leaders and provide their honest opinions on state initiatives when they fear that their arguments could have adverse consequences. For instance, regime leaders may be open to direct criticism in a closed forum, but may fear the damage such criticism could do to their legitimacy if heard by a larger audience.

There is formal theoretical support for this conjecture. Andrea Prat (Reference Prat2005) has demonstrated that when the outcomes of an agent's acts are observable (e.g., televised news reports of activities in a national parliament), but the relationship between actions and outcomes is better understood by the agent (legislator or politician) than the principal (voter) (i.e., understanding of how policy pronouncements made in parliaments are shaped into legislation), more transparency could actually lead the agent to engage in conformist behavior and refrain from actions that could yield better outcomes. Prat argues that this is the motivation behind executive privilege in the United States and other countries, where it is feared that open access to discussions by those counseling key decision makers may damage their ability to be frank and candid. As evidence of this effect, Naurin (Reference Naurin2007) shows that transparency reforms indeed had a conforming effect on representatives at the European Council, who feared that the negotiations between lobbyists and politicians would become public. In fact, policymakers are keenly aware of this problem. Countries that have adopted open information codes specifically separate information about the decision-making process and the final decision. In most countries, citizens cannot request the working papers and recommendations underlying a public decision until after the decision has been implemented, so that the leaders can receive honest counsel during the decision-making process (Frankel Reference Frankel2001).

Prat was primarily interested in how transparency limited the candid information provided by corporate boards to chief executives, but his adverse consequences hypotheses is a helpful guide to considering when and how transparency will be effective more generally. In settings where agents (legislators) understand the complex art of policymaking better than the principals (voters) and feel the need to engage in types of negotiations that would be publicly embarrassing, transparency may lead to less effective activity.

If the co-optation theorists are right, there is a critical difference in the way we should conceptualize the behavior of legislators in democratic and in authoritarian parliaments. Adopting the Prat language, in a democratic legislature, the legislator is the agent and the voter the principal. Thus, the actions taken by the legislator are meant to be observed directly and responded to by voters. As noted earlier, transparency in a democratic setting can have socially beneficial or perverse effects, such as political pandering, although the empirical literature has tended to show that the incentive and selection effects of transparency overpower the negative responses. Regardless of what actions the legislator takes, however, in democracies there is thought to be a direct signal between politician and voter.Footnote ²

We argue that in an authoritarian assembly, as described by the co-optation literature, the signaling relationship becomes more complex. In the messy and secretive world of authoritarian bargaining, concessions to potential opposition requires a delicate dance in which authoritarian leaders must be made aware of the demands of the group and benefit from alternative sources of information, but leaders fear that agreeing openly to such demands may inspire other groups, or may reveal regime weakness to the population or potential threats. Under these settings, co-opted delegates, like their democratic counterpart, sends a signal through their legislative actions, but the key recipient of the signal is the authoritarian ruling party or dictator. The voting public under an authoritarian regime is relegated to an observer of the signaling relationship between the legislator and the ruling party. Voters play a far less important role in how delegates will actually behave, but their understanding of activities in the legislature still has important implications for the specific actions taken by delegates. To maintain regime stability, the authoritarian leadership would prefer the voting public to believe that the preferences of the legislators and the regime are aligned.

Legislators, in order to retain their connections to top leaders and opportunities for promotion to government positions, as well as avoiding punishment for breach of the implicit agreement, have an incentive to make sure their behavior in an authoritarian parliament is in line with the expectations of the ruling party/dictator. As a result, they will make an effort to participate frequently and provide honest advice to the leaders, which provides crucial information to authoritarian regimes on policy choices and decisions that will help them maintain power. In a closed or shielded setting, this advice may be quite critical, but will nonetheless provide useful information. For instance, delegates may talk about dissatisfaction with socioeconomic policy in their provinces that, if unaddressed, could lead to greater problems or unrest. Both citizens and leaders benefit from this exchange. As a result, in a nontransparent authoritarian assembly, delegates will be rewarded both for the level of effort they display and for the quality of the information they provide.

Opening up a legislative session to transparency under authoritarian conditions, however, could undermine this important channel of information. Because legislators must worry about how their statements are observed by voters and whether or not they appear in line with regime policy, their ability to provide honest advice to top leaders will be compromised, and they are likely, as Prat (Reference Prat2005) put it, to behave in a conformist manner. In the case of Vietnam, where 70% of delegates do not speak during query sessions, conformist behavior implies not asking any questions that might criticize or embarrass top leaders. Thus, shining a light on legislative activity may actually force potential opposition to curtail their demands and leaders to remain more recalcitrant in their concessions.

Conformist behavior, however, can be logically separated from the level of participation or effort. Because dictators will want delegates to publicly demonstrate that the assembly is aligned with the regime, delegates may increase legislative effort that is not critical of regime leaders in the presence of transparency. However, at the same time, they may limit the amount of dangerous information provided in their legislative sessions. Our empirical analysis, therefore, seeks to separate “quality and sensitivity” of information from “nonsensitive effort” in order to understand the true impact of transparency. Therefore, we divide the incentive hypothesis, proffered for democratic states in the preceding, into two alternative hypotheses for authoritarian legislators:

These hypotheses apply solely to the incentive for changes in delegate behavior, but it is important to consider the impact of transparency from the perspective of the authoritarian leader as well. Because the authoritarian leader is worried about regime stability, a transparency intervention also has the effect of raising the value of the delegate's behavior to the leadership. In other words, transparency increases the authoritarian regime's incentive to buy a particular delegate's silence on sensitive or destabilizing issues. The price of silence may be public goods for the local district, promises of renomination to the VNA, or career advancement. Under conditions of transparency, leaders also have an incentive to buy legislators' proactive support and endorsement on all other issues. In sum, we expect that participants in a co-optive exchange might be rewarded for living up to the relationship (i.e., through effective query participation and information provision). Contrariwise, delegates should be punished by authoritarian leaders when they receive the benefits of the arrangement (e.g., service in the VNA and role in policymaking), but fail to live up to the terms of the agreement (i.e., use their position in a manner that might threaten regime stability). Hence, in contrast to democratic systems, we propose two selection hypotheses that acknowledge the power of central authoritarian elites:

Although the conduct and results of authoritarian elections are often controlled or manipulated by the ruling elites, there is wide variation in the competitiveness of authoritarian elections and the role that voters can play in altering results (Gehlbach and Keefer Reference Gehlbach and Keefer2011; Hyde and Marinov Reference Hyde and Marinov2012). For our experiment, it is important to note that Vietnam has universal elections and candidate-to-seat ratios in electoral districts that range between 1.5 and 2, implying that there is a possibility that delegates can lose. Indeed, previous research has indicated that before the experiment there was a strong and robust relationship between delegate activity and the competitiveness of elections in their district (Malesky and Schuler Reference Malesky and Schuler2010). The vote share a delegate received in the 2007 election was significantly and negatively correlated with delegate participation, criticism, and representation of constituency in the query sessions.Footnote ³ The strong relationship between vote share and activity is important because it offers the intriguing possibility of responsiveness to voters in an authoritarian regime. In short, even in an authoritarian setting, voters can still respond to candidates, by turning out and voting, or by protesting through spoiled ballots. Whether voters decide to respond or not depends on two factors: (1) how much they think their responses matter for the election outcome; and (2) the activity of the delegate in the legislative session. It is possible that voters may reward candidates who represent their interests in query sessions and legislative debates. To capture this, we propose the following voter responsiveness hypothesis:

Participation in Legislative Debates

Legislative debates allow delegates to put forth ideas on specific draft laws. These ideas are then compiled and reported to the Standing Committee of the VNA. If the law is to be decided on during that session, the Standing Committee reports these comments to a drafting committee for consideration before the final version of the law is drafted. For laws to be decided in upcoming sessions, the ideas are given to the drafting committee for review before the law is submitted to the VNA for a vote in future sessions.

Thus, all comments during the debates are related to a specific law or report. In the June 2010 session, the full assembly convened to consider 16 bills, which included the Law on Food Safety and the Law on Mineral Resources. Of the bills that were debated, 10 were passed during that session, whereas six were left to be decided in a subsequent session (Voice of Vietnam News 2010). Because the debates focus on the bills, no ministers or government officials are questioned directly. Instead, the chair or one of the deputy chairs of the VNA will set the guidelines of the discussion, sometimes followed by a report by one of the chairs of the VNA's 10 substantive committees. The draft law is then opened up for comment by the delegates.

Because of the focus on legislation, speeches in these sessions demand familiarity with the subject and require significant research on the part of the delegate. A typical comment heard during the debates would be one that Hồ Thị Thu Hằng (from Vĩnh Long) offered on the Law on Food Safety:

This comment is representative of the general tenor of the debates, as recorded in VNA transcripts.Footnote ⁴ Although they can be critical of the proposed legislation, they do not focus on holding the government or specific leaders accountable. They also demonstrate that the delegates have read the laws in great detail, which requires significant effort. However, because these comments do not challenge any specific ministers, they are not evidence of the willingness of delegates to challenge the central authorities. The only exceptions are the debates on reviewing the state's performance in meeting previous socioeconomic targets as well as developing future targets. Despite these instances of direct criticism, of the 738 pages of transcripts from the June 2010 session, 642 pages were devoted to discussing specific bills with only 94 devoted to the socioeconomic targets.

Participation in Query Sessions

The VNA query sessions have been conducted twice a year since the ratification of the 1992 Constitution, which sought to give the VNA a greater role in policymaking. In contrast to legislative sessions, comments made during the query sessions directly question government officials. Of the 1,260 questions asked during the seven VNA sessions held between 2007 and 2011, only 197, or 15%, specifically mentioned a piece of legislation. Even then, the legislation was not usually mentioned in order to suggest a change to the law, but rather as evidence to hold the minister accountable. Although policy is not made during these sessions, delegates present grievances from different regions and sectors of society directly to the ministers. Sometimes the questions can become quite pointed. In an extreme case, a minister has been forced to resign, such as when Minister of Transportation Đào Đình Bình stepped down after failing to defend his role in a major corruption scandal. In 2003, a delegate requested a vote of no confidence after a particularly poor performance by the Minister of Education (Salomon Reference Salomon, Balme and Sidel2007).

Although some questions are clearly more critical than others (see Malesky and Schuler Reference Malesky and Schuler2010 for examples), even mild queries bring a degree of public pressure to bear upon the ministers in the hot seat. The following query was put forward by Cao Thành Văn (from Bạc Liêu) during the questioning of the Minister of Information and Communication:

Although this passage was coded as “uncritical” because it did not directly impugn the performance of the Minister, Ministry, or Government, it did force the minister to commit to a solution or admit that, as of yet, he did not have one. In this instance, Lê Doãn Hợp, the minister, failed to answer the questions during his first response. Then, several minutes later, Cao Thành Văn rose to reiterate his query. Hợp acknowledged that he did not have any information about the issue and called on the other VNA delegates to provide him with details on overcharging. As this exchange shows, even when the delegates are not directly critiquing cabinet members, their questions put them under pressure and can lead to public embarrassment.

Given the differences between the two types of participation in legislative sessions, it is clear that speeches in legislative debates more accurately represent delegate effort on nonsensitive activities, whereas query sessions are far more likely to play the role of benign criticism and sensitive participation expected by co-optation theorists. Legislative debates ask delegates to perform their legislative duties, requiring them to put a great deal of work into reading the draft laws and providing competent criticism of the drafts. However, in contrast to the query sessions, no central official is directly questioned during these sessions, making it unlikely that the delegates will feel any pressure from above to curtail their comments. During the query sessions, however, delegates must possess the willingness to challenge an official, even when this might put them at odds with a higher-ranking official. Thus, in our empirical analysis, we treat participation in legislative sessions as evidence of nonsensitive effort, whereas query session participation is treated as a measure of participation on sensitive activities, in order to assess H1 and H2 separately.

Given the dangers of transparency for an authoritarian regime that we discussed earlier, it is critical to examine the reasons the Vietnamese government would allow query sessions and public transcripts of legislative activity at all. The answer is best explained by contributors to the literature on power sharing in authoritarian regimes, as query sessions are an example of the arrangements they believe are so important for authoritarian survival (Boix and Svolik Reference Boix and Svolik2010; Gehlbach and Keefer Reference Gehlbach and Keefer2011; Svolik Reference Svolik2009). By giving space to criticize policy, the center can credibly claim to be allowing genuine, if limited, power sharing. By using this option, delegates can cause damage to the center, and thus have some restraining power on leadership excess. Nevertheless, the co-opted elites who exercise this option may not be punished. In equilibrium, the elite must have the option to criticize the center in a damaging way in order for the power sharing to remain credible. However, to ensure the maintenance of stability, which is the other component of co-optation, the center must have the ability to punish those delegates when criticism leaks into the public or threatens the regime. Therefore, we should see rare cases where delegates miscalculate, launch damaging attacks on the center, and are subsequently punished. But the severity of the punishment will be tempered by the need to maintain the power-sharing relationship. Before a transparency intervention, the query session fulfilled this task extremely well.

Citizens’ Knowledge of Legislative Sessions

The two-week-long sessions have attained greater prominence in the broader public eye ever since the VCP began televising parts of them in the mid-1990s. Although most Vietnamese do not watch the full sessions, many pay attention to the highlights that are shown on evening news programs and discussed in Vietnam's numerous newspapers. As a consequence, large proportions of Vietnamese claimed to be familiar with the query sessions in a recent survey (UNDP 2012).Footnote ⁵

However, highlights do not necessarily give voters a clear idea of what their delegates are doing, and do not represent a threat to regime stability, for three reasons. First, news programs tend to focus on the most interesting speeches and debates, leaving a large portion of the interactions uncovered. Transcripts of queries and responses are posted on the VNA web site after each session, but these are not presented in an easy-to-find location and have wildly differing titles. Moreover, the transcripts are posted in Word files of two hundred pages or more that make it difficult for citizens to identify quickly what their delegates said. Given that delegate votes on individual bills are not made public, the lack of a systematic account of delegate performance during the query sessions means that voters have almost no way of knowing what delegates are up to in Hanoi. Second, though some stinging criticisms make it to the media, the regime still has methods at its disposal to suppress many powerful criticisms, particularly by pressuring the state-owned media to suppress coverage of the more inflammatory delegates. Vietnamese journalists are informed on a weekly basis of which stories can and cannot be covered (Heng Reference Heng and Marr1998; Reference Heng and Hock Guan2004). It is not uncommon for particularly sensitive query sessions and performances by leading officials to be added to the list of forbidden subjects. Third, under the current method of disseminating information on VNA activity, citizens have no way to monitor the performance of any particular delegate. They only see a quick briefing of highlight reel statements.

The benefit of our treatment, however, was that it made all the information publically available in an easily accessible and measurable format. Visitors to our pages on the VietnamNet web site could select the delegates they wanted to view, read direct transcripts from the sessions, and compare their own delegates to others in the VNA using our prepared tables. Before our project, the only way to tally the performance of an individual delegate was to trawl through the hundreds of pages of publicly available transcripts of the sessions, a process that took us over a month with the aid of several research assistants.

Balance between Treatment and Control Groups

To ensure balance in our randomization, we stratified the sample frame of 462 delegates on the four variables that have been previously shown to determine who speaks in the query sessions. These variables included (1) whether the delegate was centrally or locally nominated; (2) fulltime status; (3) whether the delegate was over the age of 65 (past retirement age), and (4) the delegate's historical activity measured by how many questions they asked in previous sessions (Malesky and Schuler Reference Malesky and Schuler2010).

Table 1 shows the result of our randomization with p-values and t-scores, based on t-tests comparing means of unequal variance between the treatment and control group for each variable. Positive t-scores indicate a higher mean score for the untreated group, whereas negative t-scores indicate a lower t-score for the untreated group. The table indicates that balance was achieved on all of the stratification variables, which was expected.

TABLE 1. Balance and Descriptive Statistics

Our experiment was conducted between the November 2009 and June 2010 assembly sessions. Prior to November 2009, there were four query sessions for the twelfth VNA, conducted in November 2007, May 2008, November 2008, and June 2009. Table 1 also shows that we were able to achieve balance in terms of activity in the first five pretreatment query sessions, with the untreated delegates somewhat less likely to speak than treated delegates, but not at levels of commonly accepted significance. However, of slightly greater concern is that members of our control group were significantly more likely to be in provinces with greater central transfers and were more likely to have won with a larger percentage of the vote share in the 2007 election. These small deviations were expected, however, given the small sample size for the treatment and the fact that there are only 63 provinces, which complicates balance in covariates measured at that level.Footnote ⁸

Possible Uniqueness of the Sixth Session

An additional empirical concern with the experiment is that the treated sixth VNA session may have been fundamentally different from previous sessions for reasons beyond our experiment. Most strikingly, the 11^th Party Congress took place in 2011. It is thought to be commonplace that political activity is restrained in the year before the Congress, as politicians and bureaucrats jockey for high-level party appointments and shy away from activities that might bring negative attention to them. Qualitatively, the sixth session of the VNA actually appeared slightly more active than previous sessions, when the VNA voted to reject a resolution (20% abstain; 38% in favor; 42% against), sponsored by the Prime Minister, to build a $56 billion high-speed railway down the coast of the country. Putting aside the merits of the proposal, this was the first time that the VNA rejected a piece of legislation sponsored by either the Prime Minister or President, and was immediately dubbed historic by the participants in the debate (Ninh Reference Ninh2010).

Quantitatively, the session does not appear to be very different from previous sessions. Table 2 demonstrates that this session was well within the range of activity shown in previous sessions. In terms of quantity of activity, the sixth session ranked in the lower half; the number of questions asked averaged about 0.37 questions per delegate, but those delegates who spoke asked over three questions each. In terms of the quality of activity, the session was similar to previous sessions. The percentage of critical questions asked (31.46%), those using the name of their province (14.83%), and those employing the word “voter” or “constituency” in a query (16.11%) were not out of the ordinary relative to other sessions.Footnote ⁹

TABLE 2. Summary Statistics by Session

^a Or making at least one speech during the debates. Detailed coding rules for critical and local questions can be found at http://dvn.iq.harvard.edu/dvn/dv/emalesky in the Nodding or Needling archive.

Table 2 also helps make an important point. In the sessions before our treatment, only a third of the delegates spoke and only a third of the speaking delegates dared criticize top central leaders. When we speak of conformist behavior under the adverse consequences hypotheses, therefore, the empirical evidence indicates that delegates should conform to the median behavior, which is to remain silent.

Internet Penetration as a Predetermined and Orthogonal Measure of Intensity

The nature of our experiment allowed for the possibility that some delegates received a stronger treatment than others. This would occur as more visitors were drawn to a particular delegate's web page, thereby exposing him/her to a higher-level of scrutiny than peers in the treatment group. Under the democratic version of the incentive hypothesis, we would expect that increased exposure to the treatment would lead to more participation in query sessions. If transparency during the policymaking process curtails participation (H1), we should find the reduction to be highest on more heavily exposed delegates.

Although it is the most direct measure, judging the level of exposure to the treatment by clicks on a delegate's page is unfortunately endogenous to the quality and quantity of a delegate's query session performance. Hit counts could increase for particular delegates if readers rushed to read sensitive, critical, or even humorous statements in their entirety. To avoid confusing dosage with popularity, we interact the treatment with the level of Internet penetration in each province in 2009, as the proportion of the Vietnamese population who read online newspapers varies dramatically across Vietnam.Footnote ¹⁴ Consequently, our treatment intensity also varies across delegates, as their constituents are located in different provinces with varying levels of Internet penetration and therefore varying levels of access to their delegates’ web pages. This variation allows us to observe the effects of different treatment intensities and is an appropriate measure for our analysis, as the infrastructure and number of Internet subscriptions predate the 2010 experiment.Footnote ¹⁵ The approach sheds a powerful light on authoritarian politics, but is more speculative, as Internet penetration was not randomly assigned.

In Table 4, we provide the results of a regression of the number of total page views for a particular delegate on our measure of Internet penetration. Total page views include hits on the delegate's introductory page, interview page, and scorecard. Among treated delegates, the average page views were 6,300, with the most popular delegate receiving 60,000 hits and a few delegates receiving no hits at all. Although we display the aggregate number, our findings are robust to regressing hits on each page separately. Model 1 provides the bivariate analysis. Model 2 controls for whether the delegate agreed to be interviewed. Model 3 controls for the number of questions asked by a delegate in Session 6 to address the endogeneity of page views. Model 4 adds individual-level control variables, Model 5 adds provincial structural controls that are plausibly correlated with both internet penetration and page views (GDP per capita, population, and whether the province is a national-level city), and Model 6 drops two outliers who received exceedingly low page views and therefore have a great deal of influence on the regression line (as measured by their Cook's D statistics).

There are a few things to note about Table 4. First, in the fully specified model, the relationship between Internet penetration and page views is very strong. Each 1% increase in Internet penetration increases page views by about 1,900 hits. Thus, we can conclude that delegates in areas with higher Internet access are more likely to face greater scrutiny of their participation from friends, co-workers, and local leaders. Second, agreeing to be interviewed increased scrutiny on individual delegates. Interviewed delegates received 14,900 more hits than other treated delegates. Finally, page views alone cannot be used in a regression model, as they are clearly endogenous to the activity of delegates in the sixth session. Each question a delegate asked drew an additional 1,300 viewers to their page.

Conditional Effect of Treatment and Internet Penetration

Using Internet penetration as a measure of treatment intensity, we interact it with the treatment in Table 5. We present the two difference-in-differences analyses from before: (1) the change between the fifth and the sixth sessions and (2) the change between the delegate average in previous sessions and the sixth session.Footnote ¹⁶ We also use the same dependent variables of questions asked and share of critical queries. Three models are displayed for each dependent analysis. The first model uses only the component variables and the multiplicative interaction, the second model controls for stratification variables, and the third model allows for provincial covariates.

TABLE 5. Intensity of Treatment Effect (Interaction between Treatment and Internet Penetration)

Notes: This table depicts the results of the controlled field experiment on two different dependent variables. In this analysis, the treatment is interacted with Internet penetration to derive an orthogonal intensity of treatment effect. The dependent variables are (1) number of questions asked by a delegate; (2) percentage of queries that were critical of ministers, ministries, or the national government. Panel A uses difference in difference between Session 6 and Session 5. Panel B analyzes difference in differences between average questions and Session 6. All analyses used OLS on the first differences between sessions. Robust standard errors, clustered at provincial level, are shown in parentheses. The first model in each set is unadjusted, the second model controls for individual-level stratification variables, and the third model controls for provincial covariates that are potentially associated with Internet penetration.

*** p < .01, ** p < .05, * p < .1

Control Variables

Although our treatment is randomly assigned, Internet penetration is not, leading to the possibility that our results could be an artifact of omitted variable bias. Consequently, it is important to control for variables that may be associated with both Internet penetration and increased (or decreased) participation in the VNA. These include (1) a dummy variable for whether the locality is designated as one of Vietnam's five national-level cities, and therefore a populous, dense, and technologically advanced urban center; (2) GDP per capita, to capture provincial wealth; (3) population, measured in millions of citizens; (4) transfers as a percentage of locally produced revenue, which gauges the dependence of a province on central beneficence and has been shown to affect delegate participation (Malesky and Schuler Reference Malesky and Schuler2010) and local-level market orientation (Malesky and Taussig Reference Malesky and Taussig2009); and (5) the quality of provincial governance, as measured by the Vietnamese Provincial Competitiveness Index, an annual ranking of economic governance for private sector development in the country, measuring such issues as corruption, property rights protection, and local transparency (Malesky Reference Malesky2009).