Category Archives: Innovative Methodology

Election Frauds, Postelection Legal Challenges and Geography in Mexico

Post developed by Yioryos Nardis in coordination with Walter Mebane.

ICYMI (In Case You Missed It), the following work was presented at the 2015 Annual Meeting of the American Political Science Association (APSA).  The presentation, titled “Election Frauds, Postelection Legal Challenges and Geography in Mexico,” was a part of the session “Detecting and Concealing Patterns in Data” on Saturday September 5th, 2015.

Political Science and Statistics Professor and Center for Political Studies faculty member Walter Mebane previously examined electoral fraud in Russia. Professor Mebane, in collaboration with Research Assistant Jonathan Wall, now turns his focus to Mexico and the Presidential elections of 2006 and 2012.

This new research by Mebane and Wall investigates if the numbers of casillas (i.e. ballot boxes) and votes challenged and nullified in Mexico reflect political strategies or genuine election irregularities. In the 2006 elections in Mexico, nullification petitions by runners up were filed against 21% of ballot boxes (27,109/130,788), even though only .56% of votes (237,736/41,791,322) were actually nullified. Similarly, in the 2012 elections, 22% of ballot boxes (32,151/143,132) were challenged and .38% of votes nullified (184,725/49,087,466).

Mebane and Wall examine how the types of nullification claims relate to ballot-box level measures of election fraud and whether the reasons cited for the challenges are uniform across the two elections and/or geography. That is, are complaints geographically clustered or does a complaint depend on geographically clustered frauds?

Ballot-box level measures of election fraud are estimated using casilla vote count data. Hotspot analysis is used to show how nullification petition challenges to casillas are distributed across geography. This technique identifies which locations have local means that are higher than the overall average values and which have local means that are lower than the overall average. A redder color indicates a cluster of locations with higher than average values, and a bluer color indicates a cluster of locations with lower than average values. Grey indicates a cluster of locations that does not differ significantly from the overall mean.

Figure 1 represents nullification complaints of type corresponding to willful misconduct or error in the vote count in the two Presidential elections, and Figure 2 represents incremental fraud probabilities for nullification complaints. Comparing Figure 2 to Figure 1 indicates that the pattern of geographic clustering for the incremental fraud probabilities does not correspond well with the pattern for nullification complaints.

Figure 1: Nullification complaints of type corresponding to willful misconduct or error in the vote count, Mexico, 2006 and 2012 Presidential election Casillas.
Seccion (precint) geographic cluster hotspots

Figure 1aFigure 1b

Figure 2: Incremental fraud, Mexico, 2006 and 2012 Presidential election casillas.
Seccion (precint) geographic cluster hotspots

Figure 2aFigure 2b

In 2006 there are more widespread regions with clusters of casillas having above average frequencies of complaints than there are regions in which there are clusters of casillas with above average incremental fraud probability values. Some of the above average type clusters overlap with above average incremental fraud clusters, but more than half do not. In 2012 on the other hand, we observe the opposite. Clusters of casillas with above average incremental fraud probabilities are much more prevalent than are clusters of casillas with above average frequencies of complaints.

Such patterns indicate that it is unlikely that the relationship between incremental fraud probabilities and the incidence of complaints are positively related. This therefore suggests that the occurrence of nullification petitions is related to the strategic and tactical incentives of political parties.

To read the full paper please visit: https://drive.google.com/open?id=0By8J0EDg6IC3MDgxOWlUVGdfakE

How accurate is marketing data?

Post developed by Katie Brown and Josh Pasek.

Photo credit: ThinkStock

Photo credit: ThinkStock

Have you noticed how the products you look at online seem to follow you from site to site and the coupons you receive in the mail sometimes seem a little too targeted? This happens because a set of companies are gathering information about Americans and merging them together into vast marketing databases. In addition to creating awkwardly personal advertisements, these data might be useful for researchers who want to know about the kinds of people who are and are not responding to public opinion surveys.

But before marketing data are incorporated into social science analyses, it is important to know how accurate the information actually is. Indeed, there are many concerns about consumer data. It could be out of date, incomplete, linked to the wrong person, or simply false for a variety of reasons. If we don’t know when marketing data are accurate, it is going to be difficult to figure out how these data can be used.

This is where the work of Josh Pasek, Center for Political Studies (CPS) Faculty Associate and Assistant Professor of Communication, comes in. Pasek, along with S. Mo Jang, Curtiss L. Cobb, J. Michael Dennis, and Charles DiSogra, have a forthcoming paper in Public Opinion Quarterly about the utility of marketing data. With Gfk Custom Research, 25,000 random addresses were selected, with about 10% of those joining the study. The marketing data available on these individuals was then matched against data collected as part of the study.

Interestingly, many variables showed large discrepancies between the two sources. Incomes mismatched by more than $10,000 for 43% of participants, while education level differed in at least two measures for 25%. Even the number of people living at the address differed by two or more in 35% of cases. Pasek and colleagues also investigate missing data with three different analyses. Ultimately, they find that the amount of data missing from consumer data is vast.

But at the same time, the consumer data performed better than chance in predicting actual data for all variables. This may make them useful for marketing purposes, but Pasek cautions that social scientific applications could be problematic. As Pasek says, “The bottom line is that these data are not consistently accurate. Although they may be great for targeting people who are more likely to buy a particular brand of shoes, our results suggest that marketing databases don’t have the precision for many research purposes.”

Significant moments throughout the history of ANES

Post developed by Katie Brown in coordination with Darrell Donakowski.

ANES65th

This is the last in a series of posts celebrating the 65th anniversary of the American National Election Studies (ANES). The posts have sought to highlight some of the many ways in which the ANES has benefited scholarship, the public, and the advancement of science.

 

As part of the 65th anniversary of the American National Election Studies (ANES), the ANES team created an interactive timeline. The timeline charts the history of the project with annotated notable dates and historic photographs. Here, we highlight three of the many entries.
Screen Shot 2014-12-11 at 11.57.17 AM

1948 – The Very Beginning

The timeline begins with the inception of ANES. In 1948, social psychologists Angus Campbell and Robert Kahn and the University of Michigan’s Survey Research Center (SRC) surveyed the national electorate. The 1948 survey served as a pilot study for, and many consider to be the first implementation of, the ANES.

1964 – The Feeling Thermometer

The 1964 wave of the study pioneered the feeling thermometer. This unique question format asks respondents to gauge their feelings on a scale from “cold” to “warm”. Feeling thermometers have since been included in all ANES waves, with their use spreading globally and to all academic fields

1996 – Comparative Study of Electoral Systems

The independent Comparative Study of Electoral Systems (CSES) project coordinates the inclusion of common sets of questions in post-election surveys around the world.  ANES first incorporated CSES questions in its 1996 wave, moving from national barometer to global participant in the process.

Please consider further exploring the interactive timeline to be reminded of some of the many significant moments throughout the history of this important scientific resource.  And if you have ideas for additions to the ANES timeline, the study team would welcome your suggestions by email to: anes@electionstudies.org

 

Visiting Scholar Profile: Khalil Shikaki

Post developed by Katie Brown.

Growing up in Palestine during a period of political tension, Khalil Shikaki became fascinated with politics. After completing a BA in Political Science at the American University of Beirut, Shikaki pursued these questions with a PhD in Political Science from Columbia University. Two core questions drive his research: What direction is Palestinian society taking? How do Palestinians view the relationship with Israel?

After finishing in 1985, he began teaching at An-Najah National University in the Northern part of the West Bank. Shikaki designed and taught his first course (the same course he is teaching here this semester): Palestinian Politics & Society. At that time, he realized there were no data to support and test theories. Even politicians relied on word of mouth to gauge public opinion.

So Shikaki set out to collect this data with Palestine’s first survey. But he faced a major roadblock. At the time, Palestine was under Israel occupation, and the Israeli military forbid the proposed survey. After the Madrid Peace Conference in 1991, things started to calm, which created an opening for data collection. In 1992, Shikaki started training survey administers and opened the Palestinian Center for Policy and Survey Research (PCPSR) in Nablus (PCPSR moved to Ramallah in 2000). And the first survey ran in 1993. Results of the survey, which explored public attitudes toward the Oslo agreement, were published in the same day the agreement was officially signed in Washington DC.

If the first challenge to data collection was Israeli occupation, the second was building trust with Palestinians. In particular, Shikaki and his team had to convince citizens that they could trust the interviewers enough to speak their mind. The initial response rate to a pilot study came in at just 50%. The team underwent additional training over the course of a year to project and inspire confidence and did not enter the field officially until this dropped to 10%. Once out in the field they found the opposite of their initial fear: if anything, people wanted to talk too much.

The surveys rolled out in both the West Bank and Gaza. When Hamas took over Gaza in 2007, the researchers faced some new issues as Hamas now wanted to control the data. But the survey continued. And while some field workers have faced arrest, they are usually released the same day.

Data collection also becomes difficult when battles rage with Israel. This past summer, the survey stopped in areas under bomb and rocket attack. But the day before the ceasefire, interviewers tackled the embattled regions. Aside from timing, the researchers also had to adjust for areas now decimated by war, citizens displaced. This allowed for timely and valuable data on opinions about how the latest fighting impacted Palestinian views.

With more than 20 years experience investigating his driving questions — What direction is Palestinian society taking? How do Palestinians feel about the relationship with Israel? – Shikaki sees a few trends. Over the first ten years, the Palestinian public was moving in a politically moderate direction with greater support for diplomacy and compromise with Israel. The most recent decade, though, has witnessed greater support for Islamists and violence with Israel. The fate embodied in these questions are linked. Shikaki remains hopeful, as a change in either would enact change in the other.

We are extremely pleased to have Khalil Shikaki in residence at the Center for Political Studies (CPS). During his time at CPS, he has also been continuing his collaborations with the Arab Barometer, the Aggression Research Program, and Scott Atran.

Measuring Political Polarization

Post developed by Katie Brown and Shanto Iyengar.

The inaugural Michigan Political Communication Workshop welcomed renowned political science and communication scholar Shanto Iyengar from Stanford University. Iyengar presented a talk entitled “Fear and Loathing across Party Lines.”

Iyengar began by considering the current polarized state of American politics. Both parties moving toward ideological poles has resulted in policy gridlock (see: government shutdown, debt ceiling negotiations). But does this polarization extend to the public in general? To answer this question, Iyengar measured individual resentment with both explicit and implicit measures.

Iyengar1

2008 ANES: Party vs Other Divisions

 

For an explicit measure, Iyengar turned to survey evidence. The American National Election Studies (ANES) indeed illustrates a significant decline in ratings of the other party based on feeling thermometer questions. Likewise, social distance between parties has increased over time, as measured by stereotypes of party supporters and marriage across party lines. In fact, this out-group animosity marks a deeper divide than other considerations, even race (see graph below).

But these surveys gauge animosity at the conscious level. Iyengar also believes mental operations concerning out-party evaluations occur outside of conscious awareness. So, along with Sean J. Westwood, Iyengar pioneered implicit measures of out-party animosity. Specifically, Iyengar and Westwood adapted the Implicit Association Test— originally used to capture racism – to political parties. Interestingly, the IAT also captured this animosity, although the polarization was more pronounced with the explicit survey measures. The chart on the left shows the starker divide between Democrats and Republicans using the feeling thermometer; the chart on the right shows the difference with the IAT.

Iyengar2

Comparing Implicit with Explicit Affect

Iyengar also adapted classic economic games to test implicit out-party animosity. Both games allow the participant to share a proportion of money provided by the researchers. Interestingly, participants gave less to out-party opponents. Iyengar cites this as evidence of implicit out-party bias.

Iyengar3

Economic Game Results by Party

Together, these results suggest marked party polarization. The hostility is so strong that politicians running on a bipartisan platform are likely to be out of step with public opinion.

Ghost-House Busters: Response to an Italian Tax Evasion Program

Post developed by Katie Brown and Ugo Troiano.

In Italy, would be tax evaders hide money in houses. In response, the Italian government developed a “Ghost Buildings” program which used technology to identify and monitor buildings otherwise hidden from tax authorities. In total, the Italian government identified two million houses that were hidden for tax purposes.

How do voters respond to anti-tax evasion policy? Answering this question has been a challenge because tax evasion is illegal and is typically very hard to measure. However, the Ghost Buildings program offers a unique opportunity to investigate the issue. Center for Political Studies (CPS) Faculty Associate and Department of Economics Assistant Professor Ugo Troiano seized this chance in a working paper along with his co-author Lorenzo Casaburi who is a Postdoctoral Fellow at the Stanford Institute for Economic Policy Research (SIEPR).

The authors approached the research knowing that anti-tax evasion programs like the Ghost Building program hurt tax evaders while benefiting tax payers. Usually, one assumes evaders to be the minority and payers to be the majority. If this is the case, a program like Italy’s Ghost Buildings one should boost public opinion in support of incumbents as payers support the crackdown on evaders.

To test the effect in Italy, the authors mapped the identified Ghost Buildings to calculate a town-by-town concentration, as illustrated by the graph below.

Ghost Building Concentration

italymap

The authors also collected geographic and socio-economic data for the same towns. They found tax evasion to be more prevalent in cities, possibly because cities offer more opportunity to to buy houses from a larger stock. Tax evasion was also found to have an inverse relationship with social capital and income.

Troiano and Casaburi then looked at the impact of the Ghost Buildings program on vote choice. Specifically, they were interested in change in support for incumbent politicians. The graph below displays the results:

Difference in Reelection Rates Pre- to Post- Ghost Buildings Program

graph

The positive trend-line shows that the higher the proportion of Ghost Buildings in a given town, the more that vote support for the incumbent increases. The authors therefore concluded that, “Local incumbents are shown to obtain positive political returns, namely, an increase in their reelection likelihood, from the Ghost Building program.”

The authors found that fighting tax evasion allowed politicians to increase public good provision, thereby creating material benefits to society. Second, support for the incumbent who fights tax evasion was found to be stronger in regions with attitudes more hostile to evaders. The authors also looked at responses to the World Values Survey question “Do you think tax evasion is justifiable?” and what they found suggests that tax enforcement and attitudes toward evasion have a complementary relationship.

Political Ads, Emotional Arousal, and Political Participation

Post developed by Katie Brown and Kristyn L. Karl.

It’s election time again. And elections bring advertising assaults by Internet, radio, and TV. In Michigan and Iowa, there is one political TV ad every two minutes. But what effect does this have on potential voters?

Center for Political Studies (CPS) affiliate and Ph.D. Candidate in Political Science at the University of Michigan Kristyn L. Karl investigated this question. Where previous research in this area uses self-reported measures of emotional response, Karl tackled the issue with a randomized experiment capturing a direct measure of physiological arousal – skin conductance. She was interested in the impact of emotional arousal from political ads on citizens’ intention to participate in politics.

Sample Skin Conductance Output

Sample Skin Conductance Output

For the study, Karl brought participants into the lab and measured their skin conductance while watching a political advertisement. The ad was fictitious and created in a way that gave Karl control over the message, images, music, and structure. Karl designed four ads: a positive Democratic or Republican ad, and attack ads on Democrats or Republicans. Participants randomly watched one of the four ads while their physiological arousal was captured; after the ad, they reported their current emotions and their willingness to participate with regard to 1) signing a petition, 2) initiating a conversation on a political topic, and 3) attending a meeting, rally, or demonstration.

Karl finds some key differences between political novices and more experienced participants. For political novices, both physiological arousal and self-reported negative emotion positively predicted participation in politics. Among political experts, however, the connection between arousal, self-reported emotion, and intended participation is more muted. Specifically, while the trend is still positive, the effect fails to reach statistical significance.

The Marginal Effect of Physiological Arousal on Political Participation by Political Sophistication

graph

Karl turns to theory to explain the limited effect of arousal on intention to participate among experts. Experts have a well-developed cognitive network about politics which, for better or worse, allows them to more easily interpret and condition their emotional responses to political stimuli. Political novices do not have this expansive network and so react in a more instinctual way. The model below captures this:

flowchart

This experiment highlights the importance of using alternative measures of emotional arousal as a complementary tool to self-reported measures. Moreover, it draws attention to the question of for whom political ads are motivating and how do they work.

Measuring and Catalyzing Change from Within: the Arab Democracy Index

Post developed by Katie Brown in coordination with Khalil Shikaki.

Can measurement promote democratization in the Arab world? Khalil Shikaki, visiting scholar at the Center for Political Studies (CPS), believes the answer is “yes.” In 2006, he set out to create an instrument to measure both the direction and sustainability of the transitional process. The resulting Arab Democracy Index is a joint project between the Palestinian Center for Policy and Survey Research (PCPSR), the Arab Reform Initiative, and the Arab Barometer.

The Arab Democracy Index is unique in that it comes from within the Arab world to reflect local experiences. It also pulls from three main sources: data on government actions, reviews of legal and constitutional text, and public opinion data. Local teams in nine to twelve countries collect this data, tailoring the standardized process based on their local expertise. Together, these sources offer insight into 40 indicators of democracy.

The Arab Democracy Index indicators break down into two types. First, there are Means, e.g., legislation, which speak to democracy de jure. Second, there are Practices, e.g., elections, which speak to democracy de facto. Data in each category tally to a total. Scores below 400 constitute an undemocratic state, 400 to 699 indicate early signs of transition, 700 to 1,000 highlight visible progress toward democracy, and scores above 1,000 pertain to already democratic nations. The graph below displays the results for all Arab countries. Overall, Means remain relatively constant while Practices show signs of improvement.

Screen Shot 2014-10-06 at 4.25.24 PM

The data can also be broken down by country, as in the graph below. As we can see, some nations are driving this positive trend while others are moving away from democracy during this period.

Screen Shot 2014-10-06 at 4.27.53 PM

The results led to reports, issued in 2008, 2009, and 2010, that not only document opinion but offer policy recommendations for policy change tailored to each country. Shikaki notes that while political and civil rights have improved, more must be done. Specifically, he recommends a focus on reforming education, social justice, and socio-economic reforms. The Arab Democracy Index also underscores an important larger point: external pressure from the U.S. Department of State can help change democracy in theory, but change in practice must come from within.

A Different and Arguably Better Understanding of Rwandan Violence

Post developed by Katie Brown and Christian Davenport.

5778975_origWho did what to whom in 1994 Rwanda? This is the central question driving the GenoDynamics project directed by Center for Political Studies Faculty Associate and Professor of Political Science, Christian Davenport, and former CPS affiliate and current Dean of the Frank Batten School of Leadership and Public Policy at the University of Virginia, Allan Stam.

Last week, Davenport updated the associated website to provide new, unreleased data that the project collected, documents that can no longer be found on the topic, and new visualizations and animations of collected data.

Also added to the website was a page specifically dedicated to a recent BBC documentary, Rwanda’s Untold Story, that features the research.  The documentary premiered on October 1, 2014 in Europe and is now available for viewing on the Internet. The film itself has prompted some controversy. The most vocal critics call those involved with the documentary “genocide deniers,” which by Rwanda law classifies as anyone who completely denies or seeks to “trivialize” or reduce the number of Tutsi victims declared by the government. Others have protested outside the BBC headquarters in London. Still others have praised the film for bringing forward a story that they felt was long overdue.

The GenoDynamics website features all of this criticism. But it also offers a glimpse into what the researchers found and how they found it.

Davenport and Stam knew that Rwanda 1994 was a time of wide-spread violence when they began investigating in depth in 2000. But they did not know “who was engaged in what activity at what time and at what place.” With funding from the National Science Foundation (NSF), the researchers content analyzed and compiled data from the Rwandan government, the International Criminal Tribunal on Rwanda (ICTR), Human Rights Watch, African Rights, and Ibuka. These sources were used to create a Bayes estimation of the number of people killed in each commune of the country for the 100 days of the genocide, civil war, reprisal killings and random violence. Davenport and Stam interviewed victims and survivors as well as perpetrators in Rwanda, and they surveyed citizens in the town of Butare. Finally, through a triangulation of information from the Central Intelligence Agency (CIA), a Canadian Military Satellite image, and Hutu and Tutsi military informants through the ICTR, they created variables concerning troop movement and zone of control. This allowed them to see who was responsible for killings in the different locations.

The work is controversial in many respects – including the degree of transparency involved, as Davenport and Stam are the only project that has made all relevant data publicly available – but the biggest controversy concerns how they challenge popular understanding. At present, the official story is that one million people were killed by the extremist Hutu government and the militias associated with them, with most of (and in some stories all of) the victims being Tutsi (upwards of 800,000 in some estimations). But Davenport and Stam found that in 1991 (according to the Rwandan census as well as from population projections back from the 1950s) only 500,000 Tutsis lived in Rwanda. Davenport and Stam further concluded that approximately 200,000 Tutsis were killed, as it was reported by a survivors organization that 300,000 Tutsi survived. While this number is less than the official number, it still represents the partial annihilation of the Tutsi population, which includes genocide but likely other crimes against humanity and human rights violations as well. But the estimation also changes the official story: the results of this research suggest that the majority of those killed in 1994 were in fact Hutu.

After 14 years of research, Davenport and Stam believe that there were several types of political violence occurring in Rwanda in 1994. The table below summarizes the different types of violence that were potentially involved (by perpetrator and victim). The larger project is trying to sort Rwandan political violence into each cell, which is incredibly difficult but useful for understanding exactly what happened.

Screen Shot 2014-10-20 at 9.48.07

The current controversy is not a new one. When Davenport and Stam presented their findings to the Rwandan government, they were told that they would not be welcome to return. When they presented their findings at the 10th and 15th anniversaries, they received more criticism. At no point was any new evidence or data provided which countered their narrative. In addition to the documentary, Davenport and Stam are working on a peer reviewed journal article and a book for a broader audience.

Maternity Leave, Pronoun Use, and Gender Discrimination

Post developed by Katie Brown and Ugo Troiano.

dv780037

Photo credit: ThinkStock

Around the globe, the average maternity leave is 118 days, but with a lot of variation. Maternity leave periods range from 45 days in Bahrain and the United Arab Emirates to 480 days in Sweden. Why the difference? And, moreover, does the variation relate to gender discrimination?

Assistant Professor of Economics and Center for Political Studies (CPS) faculty associate Ugo Troiano, along with Yehonatan Givati of The Hebrew University of Jerusalem Law School, investigated these questions and published their results in an article in The Journal of Law and Economics.

Their core theory is simple: maternity leave is costly to employers, so when the government mandates long periods of maternity leave, women may be discriminated against and receive lower wages as a result. However, places with a low tolerance for pay gaps won’t let that happen, and the incidence of this policy ends up being shared across genders. As a consequence, it becomes more economically “optimal” to enact longer mandatory maternity leaves in societies that are not discriminatory toward women. This theory is consistent with the empirical evidence.

The authors show that maternity leave is indeed shorter in societies where survey respondents think that men should receive preferential treatment in the workplace. The authors measure this attitude through a World Values Study question that asks level of agreement with, “When jobs are scarce, men should have more right to a job than women.” Aggregated to the national level and combined with data on the length of maternity leaves, they find the two are related.

However, that evidence alone does not allow us to conclude that culture is causing a longer leave, because it’s possible that causation goes in the opposite direction: a longer leave is causing more tolerant attitudes toward women. In order to alleviate this concern, Givati and Troiano looked at a proxy from attitudes less likely to change because of maternity leave: language. Specifically, they developed a measure from recent research in psychology and linguistics known as linguistic relativity suggesting that language shapes thought. Following from this, they analyzed the number of gender-differentiated pronouns across languages to see if this corresponded to tolerance of gender discrimination. For example, English uses one gender-differentiated pronoun associated with gender and the number of people referenced, while Spanish uses three (see the table below). In all, Troiano and Givati tallied the number of gender-differentiated pronouns in 33 languages using grammar books, and they looked to see whether the number of pronouns was negatively correlated with gender tolerant attitudes.

table

The authors indeed find a connection, with more gendered pronouns correlated with greater tolerance of gender discrimination as measured by the World Values Survey question. Next, the authors estimated a regression model, with length of maternity leave as the dependent variable and number of gendered pronouns as the independent variable. Troiano and Givati find that the more gendered pronouns, the shorter the maternity leave. Or, on the flip side, the fewer gendered pronouns, the longer the maternity leave.

Troiano concludes, “This project is an example of how concepts from different social sciences can be used to answer specific policy questions. To understand the factors that affect the variation of maternity leave across countries we integrated economic concepts (incidence of a policy), sociological ones (measuring attitudes from survey) and linguistic ones.”