Politics is local: State legislator voting on restrictive voter identification legislation

Transcription

1 589804RAP / Research & PoliticsMcKee research-article2015 Research Article Politics is local: State legislator voting on restrictive voter identification legislation Research and Politics July-September 2015: 1 7 The Author(s) 2015 DOI: / rap.sagepub.com Seth C McKee Abstract The marked increase in restrictive voter identification (ID) laws since the 2010 elections reveals the extreme partisan polarization in those state legislatures advancing this reform. Unlike previous studies that examine state-level factors expected to influence passage of restrictive voter ID bills, this study is the first to investigate the question using the state legislator as the unit of analysis. Multivariate analysis of the voting behavior of state legislators shows which kinds of district-level factors increase or decrease their likelihood of supporting stricter voter ID laws. Given the differentiable coalitions favoring Democratic and Republican candidates, certain partisan-aligned district demographics influence state lawmaker support for restrictive voter ID legislation. Race in particular is a major cleavage conditioning support for restrictive voter ID laws. Unlike the mixed findings generated by macro-level studies, this article provides convincing evidence that the size of the black district population negatively influences the likelihood that a Democratic legislator votes in favor of a restrictive voter ID bill, but positively affects the probability that a Republican lawmaker votes yes. The findings in this study illuminate the contextual factors that influence legislator voting on this salient election reform. Keywords Voter ID laws, partisan polarization, state legislator, black voting age population Introduction The recent wave of restrictive voter identification (ID) laws has been followed by a spate of studies examining their effects on participation, as well as the motivation behind their passage. The literature has produced decidedly mixed findings on the question of how voter ID laws influence participation. Some studies display statistically significant and negative effects on turnout (e.g. Barreto et al., 2009) while others produce null results (e.g. Hood and Bullock, ). Whether or not voter ID laws are found to negatively affect participation, the contextual and measurement issues inherent in this line of research makes these examinations empirically cumbersome (see Erikson and Minnite 2009). Perhaps this explains the shift in scholarship toward understanding the factors conditioning passage of voter ID laws, irrespective of whether or not they register a measurable participatory effect. Several recent studies have looked specifically at the question of which factors influence passage of restrictive voter ID laws (Bentele and O Brien, ; Hicks et al., 2015; Rocha and Matsubayashi, 2014). Given the obvious reality that the political parties are highly polarized on this issue, with most Republicans favoring restrictive voter ID laws and most Democrats opposed, there is consensus that party affiliation is a necessary if not sufficient condition for their enactment. However, disagreement persists regarding the role that certain demographic factors, like race/ethnicity, income, education, age, and region, have toward affecting support for restrictive voter ID laws. Unsurprisingly, variation in the results of these studies is tied directly to variation in the research design and modeling of the factors expected to influence support for restrictive voter ID laws. The common thread linking these studies is that they all employ a macro-level approach, using state legislatures as the unit of analysis to assess the factors influencing support for this kind of restrictive voting measure. This article departs from the extant literature by undertaking a micro-level analysis of support for restrictive voter ID laws, examining the question of which factors influence support for such legislation when the state legislator is the unit of analysis. The findings suggest that, even though party affiliation is the overriding component driving state Texas Tech University, USA Corresponding author: Seth C McKee, Texas Tech University, Holden Hall 16, Lubbock, TX 79409, USA. sc.mckee@ttu.edu Creative Commons Non Commercial CC-BY-NC: This article is distributed under the terms of the Creative Commons Attribution-NonCommercial 3.0 License ( which permits non-commercial use, reproduction and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and Open Access pages (

2 2 Research and Politics Percent Voting Yes/No on Voter ID Law AL KS MO NH RI SC TN TX Republicans Democrats WI MN NH PA VA AR NC VA All Figure 1. Massive partisan polarization on voter ID laws in state legislatures,. legislator voting behavior, several local, district-based contextual factors register highly significant and substantively large effects on the likelihood of supporting restrictive voter ID legislation. Data and method The 2010 elections ceded Democratic control of several state legislatures to the victorious Grand Old Party (GOP). Once in command, many of these newly instated GOP legislative majorities proceeded to pass restrictive voter ID legislation. Figure 1 presents data on a sample of states that voted on restrictive voter ID legislation after the 2010 midterm election ( to ). In each state and year, the bars display the percentage of Republican lawmakers voting in favor of restrictive voter ID legislation and the percentage of Democrats voting against it. Except for the anomalous case of Rhode Island, where a Democraticcontrolled legislature introduced a restrictive voter ID measure and partisan polarization was modest (100% of GOP legislators voting yes and 69% of Democrats voting yes), in all of the other cases the GOP pushed for restrictive voter ID legislation and partisan polarization was very high absolute in Missouri and South Carolina in, and North Carolina in. Despite the partisan chasm in support of voter ID laws since the 2010 midterm, it warrants examining the factors influencing the likelihood of voting in favor of these measures. District-level data were compiled on all of the state legislators who cast a vote either in favor or against a voter ID law from the sample of states displayed in Figure 1. With the addition of four legislators not affiliated with the major parties, this consists of a total of 2400 observations. Probit regression is employed to assess what kinds of variables affect the probability of voting for stricter voter ID legislation. I construct three models: the first includes all state legislators (the entire sample), the second is limited to Republican legislators, and the third is confined to Democratic legislators. The dependent variable is coded 1 if a legislator votes yes on a stricter voter ID provision and 0 if a legislator votes no; abstention votes are excluded. After presenting the results for the three regressions, I display predicted probabilities of voting in favor of restrictive voter ID legislation based on the observed-value approach (for details, see Hanmer and Kalkan, ). A total of 14 independent variables are included in the regression with all legislators. In addition to the primary variables designating whether a legislator is a Republican (1 = Republican, 0 = otherwise) and their race (1 = African American, 0 = otherwise), there are eight district-level covariates and four other dummy controls. Data on the eight district-level variables were obtained from the Missouri Census Data Center (MCDC) website, via its American Community Survey Standard Profile Extract Assistant. This data extraction program allows one to retrieve district-level data on a host of variables constructed by the US Census Bureau s American Community Survey (ACS), which is routinely administered to a large and representative sample of the American population. The ACS district-level data are based on a five-year estimate (from 2007 to ). The MCDC includes data on state legislative boundaries both prior to redistricting (old) and after redistricting (new), which ensures that the districtlevel data match the current districts at the time a vote is cast on restrictive voter ID legislation (e.g. Virginia redrew its

3 McKee 3 state legislative boundaries before its elections and the data are based on these new district boundaries). The eight district-level variables are: % Black Voting Age Population (VAP), % Hispanic VAP, % Under 25, % Over 65 (% between 25 and 65 is the omitted comparison category), Median Household Income (divided by 1000), % Married, % Bachelors Degree or Higher, and % US Citizens. All three models include controls for region (1 = South, 0 = otherwise), legislative chamber (1 = Senate, 0 = House), and year (dummies for and, with as the omitted comparison category). In the model for all legislators, I expect that the indicator for Republican legislator should have a highly significant and positive effect on the likelihood of voting for a stricter voter ID law. By contrast, African-American legislators are expected to be significantly less supportive of voter ID legislation. Also, there is a strong expectation that the higher the district percentage Black and Hispanic VAP, the lower the likelihood of supporting a voter ID bill. Because African Americans and Latinos are important groups comprising the Democratic coalition, higher district numbers should deter legislators from supporting legislation that negatively impacts these constituents. Since the current generation of senior voters are most supportive of the Republican Party (exit poll data confirm this), it is expected that compared to the district population under 65 (and over 25), a larger senior population will have a positive and significant effect on the probability of favoring voter ID legislation. It is well known that the long-standing class cleavage in American politics divides the parties, with wealthier voters aligning with the GOP. Thus, a higher district median household income should positively affect support for voter ID laws. It is also true that on the higher end of the education scale, voters are more inclined to vote Democratic, and thus I hypothesize that a higher district percentage of college graduates reduces the propensity to vote yes on a voter ID bill. With regard to the two remaining district-level variables, % Married and % US Citizens, there are no strong expectations with respect to their effect (if any) on the likelihood of favoring voter ID laws. For the model limited to Republican legislators, because of their overwhelming support for voter ID laws, it is likely that very few variables will register an effect. One possible exception is the Black VAP. Unlike Democrats, most of whom rely on capturing a substantial portion of the African- American vote, few Republicans build winning coalitions with black support and thus the Black VAP will either register a null effect or may in fact positively affect the likelihood of voting in favor of voter ID legislation since African Americans are a critical component of the Democratic opposition and restrictive voting laws of this kind are intended to curtail Democratic participation. By contrast, in the model confined to Democratic legislators, because African Americans are a fundamental faction in the Democratic coalition, an increase in the Black VAP should reduce the probability of supporting voter ID laws. Furthermore, because the African-American population is significantly larger in the South, Democratic opposition to voter ID should be significantly higher in this region. With respect to the Senate indicator, these districts are typically larger and hence more racially diverse, making it likely that State Senators are less supportive of voter ID legislation. Finally, the year dummies should work in the opposite direction for Republicans and Democrats. Republicans should be more inclined to vote yes on voter ID after because it is an accepted electoral strategy, whereas Democrats should be less inclined to back voter ID laws after since they view these laws as ploys to suppress minority voting. The summary statistics for all the variables (mean, standard deviation, minimum, and maximum) can be found in the online appendix (available as supplementary material). Results The results for the three models are displayed in Table 1. Starting with the model that includes all legislators, the indicator for Republican lawmaker has, not surprisingly, a positive and statistically significant effect on the likelihood of voting yes on a restrictive voter ID bill. Republicans are the primary backers of voter ID laws and a host of controls only serve to reinforce how polarized voting is on these politically divisive measures. It is also true that among all legislators, African Americans are less likely to back voter ID laws. Likewise, higher minority constituencies (in terms of % Black VAP and % Hispanic VAP) reduce support for voter ID legislation, whereas higher senior populations and more affluent district residents increase the likelihood of voting yes on voter ID. A more educated district population reduces support for voter ID, and this is also true when the married and US citizen populations increase. Finally, as hypothesized, State Senators are less supportive of voter ID, and compared to, in there was a reduction among State Senators in support for voter ID laws. In the model limited to Republicans, I find that it is indeed the case that an increase in the Black VAP positively influences support for voter ID legislation. Only three other variables attain statistical significance. First, district populations with a higher percentage of Bachelors or other advanced degree holders reduce support for voter ID. Second, State Senators are not as likely as their House colleagues are to vote in favor of voter ID laws. Finally, Republicans were even more supportive of voter ID provisions in vis-à-vis, but the level of support in was not significantly different from Republican voting in. In the case of Democrats, African Americans are no more likely to vote against voter ID measures than are other Democrats. This said, because of more variation in Democratic support of voter ID laws, I find that several

4 4 Research and Politics Table 1. The likelihood of supporting restrictive voter ID legislation. All Legislators Republicans Democrats Republican 3.268*** (.120) African American.541*.057 (.284) (.380) Black VAP (%) 1.216** 5.552*** 3.078*** (.611) (1.724) (1.020) Hispanic VAP (%) 1.157* (.663) (1.089) (.986) Under 25 (%) (1.370) (2.204) (2.136) Over 65 (%) 6.389*** *** (2.187) (3.278) (4.077) Median Household.043*** *** Income (thousands) (.006) (.010) (.011) Married (%) 4.508*** *** (1.048) (1.643) (1.773) Bachelors or more (%) 4.517*** 1.634* 6.029*** (.660) (.985) (1.215) US Citizens (%) 3.001* * (1.576) (3.010) (2.323) South ** (.134) (.201) (.260) Senate.269***.356**.037 (.104) (.142) (.191).821***.377** 1.825*** (.117) (.173) (.222) ** (.153) (.270) (.366) Constant (1.593) (3.056) (2.359) Observations 2,400 1, Pseudo R Note: The dependent variable is coded 1 = vote in favor of stricter voter ID law, 0 = oppose. Probit regression coefficients with standard errors in parentheses. The number of Republicans and Democrats does not add up to 2400 because there are four legislators not affiliated with the major parties (including three independents and one Green Party adherent). State fixed effects are not included in the models because of collinearity with the South dummy and the fact that the absolute partisan polarization in certain legislative chambers predicts perfect failure for some state dummy variables. Nonetheless, omitting the South and adding state fixed effects does not alter the substantive findings. ***p <.01, **p <.05, *p <.10 (two-tailed tests). district-level factors influence this decision. As expected, opposition to voter ID increases as the district % Black VAP goes up. And similar to the model with all legislators, higher senior populations and wealthier constituents increase support for voter ID legislation, while a higher number of married constituents, highly educated constituents, and a higher number of constituents who are US citizens all reduce the likelihood of favoring voter ID bills. As hypothesized, southern Democrats are significantly less likely to vote yes on voter ID. And, based on the expectation that Democrats have generally come to perceive voter ID laws as a means to marginally advantage Republicans, Democratic support for these measures is significantly lower in and versus. Since it is not intuitive to assess the size of the effects of coefficients in limited dependent variable models like those displayed in Table 1, Table 2 converts the effects of the probit regression coefficients into predicted probabilities based on the observed-value approach (Hanmer and Kalkan, ). For each of the three models, Table 2 displays the predicted probability that a legislator votes yes on voter ID legislation for those variables that are statistically significant (p <.10, two-tailed). For the district-level variables, which are all continuous measures, the probabilities displayed include the likelihood of voting yes when each variable is set at its minimum and maximum observed value. For instance, in the model with all legislators, when the Black VAP is at its minimum observed value (.09%), the

5 McKee 5 Table 2. Predicted probability of supporting restrictive voter ID legislation. Probabilities and Differences All Legislators Republicans Democrats Republican.95,.09 [.86] African American.57,.64 [.07] Black VAP.65,.49 [.16].93,.99 [.06].17,.00 [.17] Hispanic VAP.64,.49 [.15] Over 65.56,.76 [.20].02,.40 [.38] Median Household Income.34,.99 [.65].01,.93 [.92] Married.78,.53 [.25].50,.01 [.49] Bachelors or more.75,.18 [.57].98,.86 [.12].34,.00 [.34] US Citizens.76,.62 [.14].37,.07 [.30] South.05,.11 [.06] Senate.61,.64 [.03].94,.97 [.03].57,.66 [.09].98,.95 [.03].01,.19 [.18].03,.10 [.07] Note: Table displays the predicted probability of voting for a stricter voter ID law given the maximum difference for continuous variables (the minimum observed value compared to the maximum observed value) and the first difference for dummy variables (with the probability in bold when the dummy variable = 1). The absolute value of the difference in predicted probabilities for each variable is displayed in brackets. Only statistically significant probabilities are presented (p <.10, two-tailed). All probabilities were generated based on the observed-value approach (see Hanmer and Kalkan, ). likelihood of voting in favor of voter ID is.65. By contrast, when the Black VAP is at its maximum observed value (93.2%), the probability of voting for a stricter voter ID law declines to.49. Hence the difference in the absolute probability of voting yes on voter ID based on the actual minimum and maximum value for the % Black VAP is.16; this is displayed in brackets. Similar effects are observed with regard to the % Hispanic VAP. In the case of the dummy variables, the first probability is based on when the indicator equals 1 and the second probability corresponds to the indicator set at 0. Because the dummy variable takes on its value when it is set at 1, the first predicted probability is shown in bold; the second probability, which is not bolded, presents values for when the indicator is set to 0. For example, in the model for all legislators, the predicted probabilities for the Republican indicator once again reveal the tremendous partisan polarization over support for voter ID laws. When Republican = 1, the likelihood of voting yes on voter ID is.95. When Republican = 0, the probability of backing a voter ID bill is a mere.09. The absolute difference in the probability of voting yes on voter ID according to legislator party affiliation is an extraordinary.86 (as shown in brackets). The model for all legislators also reveals that African Americans are less likely to vote in favor of voter ID by a probability of.07. At first impression it seems curious that the predicted probability that a black legislator supports voter ID is.57. But this is the likelihood when all the other controls are set at their observed values, and thus there are more Republican legislators in the dataset (pushing the probability up). There are in fact a few black Republicans, and they voted in favor of voter ID legislation. There are a total of 225 African-American legislators in the dataset who cast a vote on voter ID legislation (10 more abstained); 222 are Democrats, two are Republicans, and one is affiliated with the Green Party. Among the Democrats, 218 out of 222 voted no (98%). The two Republicans (both Texans) voted yes, and the Green Party legislator (an Arkansan) voted no. It is evident from Table 2 that, even though party affiliation has such an overpowering influence on the probability of voting in favor of a stricter voter ID law, many other variables register considerable effects. For instance, in the model for all legislators and the model confined to Democrats, most of the district-level variables have a considerable influence on the likelihood of backing voter ID legislation. Nonetheless, it is important to be aware of the standard deviation and range of values that each variable takes on (as displayed in the online appendix) since the displayed probabilities are based on the minimum and maximum observed values of each variable. For example, in the model for Democratic legislators, median household income ranges from a low of US$18,000 to a high of US$133,000. At the minimum income there is almost no chance (.01) that a Democratic legislator would vote for a stricter voter ID measure. By contrast, at the highest median household income a Democratic legislator is almost certain to favor voter ID (.93). In fact, this is the only variable in Table 2 that shows a larger overall effect on the probability of supporting voter ID than does the Republican dummy found in the model for all legislators. Given the significance of the relationship between race, party affiliation, and support for voter ID laws, a graphical presentation sheds additional light on how polarizing restrictive voter identification measures have become. Figure 2 plots the predicted probability of voting for voter ID legislation under the condition of all legislators, just

6 6 Research and Politics Probability of Supporting Voter ID Law Republicans All Legislators Democrats % 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Black Voting Age Population Figure 2. District Black Voting Age Population and support for restrictive voter ID legislation. Note: Probabilities were generated based on the observed-value approach. The maximum percentage of the Black Voting Age Population for Republicans was 35.6%, and 93.2% for all legislators and Democrats. Republicans, and only Democrats, according to the Black VAP in 10% increments. The observed minimum Black VAP for Republicans is.09% and the observed maximum is 35.6%. For Republicans the plotted values in Figure 2 range from 0% to 40% Black VAP. The observed minimum Black VAP for all legislators is.09%, a value from a Republican district; for Democratic districts only, the observed minimum is.2%. The observed maximum Black VAP for all legislators, and for only Democrats, is 93.2% (a value from a Democratic district). For all legislators and only Democrats, the plotted values in Figure 2 range from 0 to 100%. With respect to Republican legislators, the extremely high probability of voting yes on voter ID legislation at the observed minimum value for Black VAP (.09%) is.93. And at the maximum observed value for Black VAP (35.6%) in a Republican district, support for voter ID is almost a certainty at.99. As mentioned previously and shown in Table 2, for all legislators the likelihood of backing voter ID legislation is.65 at the minimum % Black VAP (.09%), whereas the predicted probability at the maximum % Black VAP (93.2%) is just.49. For Democrats, the likelihood of supporting voter ID is.17 at the minimum % Black VAP (.2%), and it declines to practically no chance of voting yes (<.01) at the maximum % Black VAP (93.2%). Figure 2 shows in stark relief just how significant the racial composition of a district is with respect to influencing the voting behavior of elected officials on an issue that sharply divides the major parties. Conclusion This is the first published study to examine voting on restrictive voter ID laws at the level of the state legislator. The multivariate models reveal several statistically and substantively significant effects on the likelihood that a lawmaker votes in favor of or against stricter voter ID legislation. Beyond the widely anticipated finding that Republicans are much more supportive of restrictive voter ID legislation, the district-level percentage of African Americans and Hispanics reduces support for these laws. This comports with the expectation that the composition of a legislator s district the local setting of politics greatly influences their voting behavior. It is also worth noting that, in alignment with the coalitional bases of support for the major parties (Green et al., 2002; Karol, 2009), among Republican legislators, a higher black district population increases legislators support for voter ID, whereas among Democratic lawmakers, a higher black district population reduces legislators likelihood of voting in favor of restrictive voter ID legislation. The relationship between the size of the black district population and lawmaker voting behavior on restrictive voter ID legislation is particularly important because it speaks to the strategic motivation behind these reforms. Whereas the connection between race and promotion of restrictive voter ID legislation is notably mixed in macro-level studies (cf. Bentele and O Brien, ; Hicks et al., 2015), in this article, which assesses the influence of district-level factors on

7 McKee 7 lawmaker voting behavior, the connection between race and support for stricter voter ID bills is robust and aligned with what we know about the coalitional bases of support for the Democratic and Republican parties. Indeed, Republican support for voter ID legislation is remarkably high, but it increases to an almost certainty when the African-American district population is over 30%. This study evaluates elites, but in this instance, the behavior of GOP lawmakers resembles that of mass-level findings on racial threat, going back to Key (1949) (see Avery and Fine, for another elite-based study of racial threat). The threat is obvious for Republican legislators: they receive so few black votes that a large percentage of African Americans could endanger their re-election bids. In this situation, the increase in Republican support for restrictive voter ID legislation is a rational response to electoral realities. The extreme partisan polarization over restrictive voter ID bills in state legislatures has not escaped the notice of the mass public. On its face, opinion on voter ID laws resembles that of a typical valence issue. However, when surveys control for partisanship, ideology, and race, the divide over voter ID laws is apparent (Gronke et al., 2015; Wilson and Brewer, ). As the issue receives significant attention from leading politicians, including the president, it appears the more attentive segment of the electorate is figuring out the correct view to adopt on the issue based on the signal they receive from elites who affiliate with their party (Carsey and Layman, 2006; Dyck, ; Dyck and Pearson-Merkowitz, 2014; Layman and Carsey, 2002; Zaller, 1992). This process of transmission can take some time to filter through the mass public (Carmines and Woods, 2002), but it appears well underway. And although a large portion of the electorate remains unaware of the reason for the stark divide among lawmakers over voter ID, elite polarization is not contingent on mass opinion. Rather, Republican legislators back restrictive voter ID laws because they expect it will marginally reduce the participation of Democratic voters, whereas Democratic lawmakers oppose these reforms because they know the reason behind their enactment. Declaration of conflicting interest The author declares that there is no conflict of interest. Funding This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. Supplementary material The online appendix is available at: by/supplemental-data References Avery JM and Fine JA () Racial composition, white racial attitudes, and black representation: Testing the racial threat hypothesis in the United States Senate. Political Behavior 34(3): Barreto MA, Stephen A, Nuno SA, et al. (2009) The disproportionate impact of voter-id requirements on the electorate new evidence from Indiana. PS: Political Science and Politics 42(1): Bentele KG and O Brien EE () Jim Crow 2.0? Why states consider and adopt restrictive voter access policies. Perspectives on Politics 11(4): Carmines EG and Woods J (2002) The role of party activists in the evolution of the abortion issue. Political Behavior 24(4): Carsey TM and Layman GC (2006) Changing sides or changing minds? Party identification and policy preferences in the American electorate. American Journal of Political Science 50(2): Dyck JJ () Racial threat, direct legislation, and social trust: Taking tyranny seriously in studies of the ballot initiative. Political Research Quarterly 65(3): Dyck JJ and Pearson-Merkowitz S (2014) To know you is not necessarily to love you: The partisan mediators of intergroup contact. Political Behavior 36(3): Erikson RS and Minnite LC (2009) Modeling problems in the voter identification voter turnout debate. Election Law Journal 8(2): Green D, Palmquist B and Schickler E (2002) Partisan Hearts and Minds: Political Parties and the Social Identities of Voters. New Haven: Yale University Press. Gronke P, Hicks WD, McKee SC, et al. (2015) Voter ID laws: A view from the public. MIT Political Science Department Research Paper No Available at: abstract= Hanmer MJ and Kalkan KO () Behind the curve: Clarifying the best approach to calculating predicted probabilities and marginal effects from limited dependent variable models. American Journal of Political Science 57(1): Hicks WD, McKee SC, Sellers MD, et al. (2015) A principle or a strategy? Voter identification laws and partisan competition in the American States. Political Research Quarterly 68(1): Hood MV III and Bullock CS III () Much ado about nothing? An empirical assessment of the Georgia voter identification statute. State Politics & Policy Quarterly 12(4): Karol D (2009) Party Position Change in American Politics. Cambridge: Cambridge University Press. Key VO Jr (1949) Southern Politics in State and Nation. New York: Alfred A. Knopf. Layman GC and Carsey TM (2002) Party polarization and conflict extension in the American electorate. American Journal of Political Science 46(4): Rocha RR and Matsubayashi T (2014) The politics of race and voter ID laws in the States: The return of Jim Crow? Political Research Quarterly 67(3): Wilson D C and Brewer PR () The foundations of public opinion on voter ID laws: Political predispositions, racial resentment, and information effects. Public Opinion Quarterly 77(4): Zaller JR (1992) The Nature and Origins of Mass Opinion. Cambridge: Cambridge University Press.