The Meaning and Measurement of Responsibility Attributions

The Measurement of Political Responsibility: Ratings Versus Rankings

Attributions of political responsibility have most commonly been measured in one of two ways. First, scholars have measured responsibility judgments through the use of ratings scales.

Tilley and Hobolt (2011), for example, invite respondents to rate government responsibility for economic conditions or health care on an 11-point scale ranging from “not responsible at all” to “completely responsible.” Similarly, Rudolph (2006) asks people to rate gubernatorial responsibility for states’ fiscal conditions on a 10-point scale ranging from “not very responsible” to “very responsible.” Focusing on the troubled aftermath of Hurricane Katrina, Gomez and Wilson (2008) employ a 4-point scale and ask people to rate whether particular political figures are responsible for “a lot of the problems, some of the problems, a few of the problems, or none of the problems.” Implicit in this ratings-based measurement approach is the belief that responsibility is a quantity that political actors possess in varying amounts. The theoretical expectation in this line of inquiry is that an actor will be held more accountable for a given outcome as his or her perceived level of responsibility for that outcome intensifies. The expectation that accountability increases with perceived level responsibility has been consistently borne out by the empirical record (Peffley & Williams, 1985; Rudolph, 2006; Tyler, 1982).

As a measurement instrument, ratings scales are effective at capturing the amount or intensity of responsibility attributed to a particular actor. When measuring attitudes, ratings scales are desirable, more broadly, for their administrative efficiency and for the cognitive ease with which they are interpreted by survey respondents (Alwin & Krosnick, 1985; Krosnick & Alwin, 1988). Importantly, though, ratings measures suffer from at least two potential limitations. Methodologically, ratings are vulnerable to response set biases and may discourage differentiation between rated actors (Krosnick & Alwin, 1988). This may occur in the present context, for example, when respondents engage in satisficing and fail to differentiate between actors by rating all of them as equally responsible for a given outcome. Ratings are potentially deficient on theoretical grounds as well because they generally fail to account for the comparative and competitive nature of attributional judgments in politics. This deficiency is particularly problematic as political elites attempt to selectively manipulate responsibility judgments by engaging in “attribution gamesmanship” and “blame management strategies” (Clarke et al., 1992; McGraw, 1991, 2001; Shields & Goidel, 1998; Weaver, 1986).

A handful of studies do attempt to incorporate the comparative nature of attributional judgments into a ratings-based approach. Peffley and Williams (1985), for example, acknowledge the comparative component of attributional judgments through the construction of their ratings scale. They invite respondents to assign blame for economic problems on a 7-point scale ranging from “President Reagan’s fault” to “President Carter’s fault.” Unfortunately, such a scale permits only binary comparisons and precludes respondents from considering multiple actors (e.g., President(s), Congress, non-governmental actors). Cutler (2008) recognizes the comparative nature of attributional judgments through the transformation of his rating scales. After asking respondents to separately rate the level of federal and provincial responsibility for a given policy outcome in Canada, Cutler (2008) calculates the relative share of responsibility by dividing a government’s rating (federal or provincial) by the combined total of both governments’ ratings (federal + provincial). Through this process, he discovers that responsibility attributions are not a zero-sum game as many respondents simultaneously attribute “above-average” levels of responsibility to multiple governments.

Alternatively, scholars have measured responsibility judgments through the use of rankings. ¹ One common practice is to present respondents with a list of actors or institutions and ask them to indicate who is “most responsible” or “more responsible” for a given outcome (Gomez & Wilson, 2001, 2003; Rudolph, 2003a, 2003b; Rudolph & Grant, 2002). Such an instrument requires respondents to make comparative judgments when deciding how to assign primary responsibility. It implicitly requires them to rank order the different actors, so that they can designate one of them as having primary responsibility. Malhotra and Kuo (2008) employ a more explicit and, arguably, more informative instrument by asking people to rank order seven different political figures according to how much blame they deserve for the lives and property lost due to Hurricane Katrina. After asking respondents “Who do you think should be blamed the most for the loss of life and property damage in New Orleans that was caused by Hurricane Katrina?” they asked who should be blamed “second most,” “third most,” . . . “seventh most?” This instrument enabled them to create a 7-point ranking in which actors could be arranged from 1 (least blame) to 7 (most blame).

When analyzing the assignment of political responsibility, a rankings-based approach has considerable merit. Although rankings are less efficient to administer and more difficult for respondents to understand than ratings (Krosnick & Alwin, 1988), they do a better job of recognizing and accounting for the comparative nature of attributional judgments in politics. In addition, they are less vulnerable to response set bias because they force differentiation between different response options (Malhotra & Kuo, 2008). Despite the fact that rankings are superior to ratings in capturing the comparative component of responsibility judgments, they do have some informational shortcomings. In particular, rankings are ill suited to provide information about the intensity of perceived responsibility. Consider, for example, a respondent who is asked to rank order the responsibility of three actors, A, B, and C, for a given outcome. If a respondent ranks A ahead of B, and B ahead of C, we know the respondent thinks that A has greater relative responsibility than B and C. What we do not know, however, is how responsible A is perceived to be in absolute terms. In the mind of this hypothetical respondent, A, B, and C could all be viewed as having either minimal or substantial responsibility for the outcome, with A deemed to have slightly more responsibility than B and C.

Toward a Theory and Measure of Comparative Responsibility Judgments

The theory underlying conventional ratings-based approaches to measuring responsibility judgments is executive centric in nature. Although not always stated explicitly, ratings-based approaches often carry an implicit presumption of presidential responsibility for the economy. The central question that voters are assumed to consider is one of intensity; that is, how much responsibility does the president bear for the nation’s economic conditions? In some respects, this is consistent with the classical reward–punishment theory of economic voting in which the incumbent, defined in terms of the president, is presumptively responsible for the nation’s economic health (Key, 1966; Kramer, 1971). Just as reward–punishment theory focuses squarely on executive accountability, the ratings-based approach seldom encourages voters to consider the responsibility of the executive relative to other governmental or non-governmental actors.

The theory underlying the rankings-based approach is not executive centric in nature. Instead, it is premised on the twin assumptions that responsibility judgments are inherently competitive and comparative and that voters are capable of differentiating between competing actors when assigning responsibility for a given outcome. The central question that voters are assumed to consider is one of comparison; that is, who is responsible for the nation’s economic conditions? The rankings-based approach is commonly grounded in selective attribution theory, which posits that voters rely on things like partisan cues and contextual information to judge the relative responsibility of competing actors (Malhotra & Kuo, 2008; Rudolph, 2003a, 2003b; Rudolph & Grant, 2002). Although theoretically richer in that it captures the comparative nature of responsibility judgments, the rankings-based approach is ill suited to provide much information about the intensity of perceived responsibility.

We argue that responsibility judgments are forged in a comparative evaluation process in which people must evaluate the relative responsibility of competing actors. We believe that a theory and measure of responsibility attributions should capture both the intensity and the comparative nature of individuals’ judgments in that evaluation process. The problem of how to capture intensity and comparability when measuring evaluative judgments has been scrutinized by students of attitudinal ambivalence. Ambivalence has been conceptualized as the endorsement of competing considerations toward a single object or the harboring of positive affect toward competing objects (Basinger & Lavine, 2005; Lavine, 2001). Ambivalence is said to be a function of the intensity and similarity of these competing considerations (Thompson, Zanna, & Griffin, 1995). To measure ambivalence, Thompson et al. (1995) proposed a computational formula, now widely used, that includes separate components to represent intensity and similarity (see, for example, Basinger & Lavine, 2005; Keele & Wolak, 2008; Lavine, 2001; Lavine, Johnston, & Steenbergen, 2012; Meffert, Guge, & Lodge, 2004; Rudolph, 2005, 2011; Rudolph & Popp, 2007). In their measure of comparative partisan ambivalence, Basinger and Lavine (2005) offer the following adaptation of the “Griffin” formula, where D represents the amount of positive affect toward Democrats and R denotes the amount of positive affect toward Republicans:

Comparative ambivalence = D + R 2 − | D − R | .

In this example, partisan ambivalence is said to be a function of “the overall intensity of the respondent’s affect toward both parties, minus the overall similarity of the respondent’s reactions toward the parties” (Basinger & Lavine, 2005, p. 173, emphasis added).

We argue that the same logic underlying such measures of ambivalence can be fruitfully applied to the study of responsibility judgments. We propose an index of responsibility in which the comparative responsibility of the president is a function of the amount of responsibility attributed to the president and the difference between the amount of responsibility attributed to the president and the mean amount of responsibility attributed to all other (J) relevant actors:

Comparative presidential responsibility = P + ( P − ∑ j = 1 J O j J ) .

The left portion of the formula captures intensity and represents the amount of responsibility assigned to the president according to a conventional ratings instrument. The right half captures comparability and represents the difference in perceived responsibility between actors. ² Comparative responsibility scores are maximized when intensity is high and when the president is judged as relatively more responsible than other actors. They are minimized when intensity is low and when the president is judged as relatively less responsible than other actors. The theoretical virtue of this measure is that it simultaneously incorporates information about intensity and comparability into a single attributional scale. ³ In this respect, it combines the strengths of both the ratings-based and rankings-based approaches into a single measure while avoiding some of their weaknesses.

Data and Measurement

Conventional Responsibility Ratings

The pre-election wave of the 2012 ANES included a battery of conventional instruments designed to measure the perceived responsibility of seven different actors for the nation’s economic conditions. Specifically, respondents were “How much is [ ________ ] to blame for the poor economic conditions of the past several years?” [President Obama, former President Bush, Democrats in Congress, Republicans in Congress, Wall Street bankers, consumer who borrowed too much money, mortgage lenders]. ⁴ Respondents were asked to rate the perceived level of responsibility for each actor along a 5-point scale using the response options of a great deal, a lot, a moderate amount, a little, or not at all. Responses are coded such that higher values denote greater responsibility.

The responses to these seven ratings instruments are reported in Table 1. As can be seen, there is considerable heterogeneity in respondents’ attributions of responsibility toward President Obama. Combining data across the top two rows, about 28% believe that Obama bears at least “a lot” of the blame. Roughly 23% feel that he deserves a moderate amount of blame whereas the remaining 50% feel that he deserves either little or no blame. When mapped onto a 0 to 1 range, the mean rating for Obama is just below the midpoint at 0.43 (SD = 0.33). Although this ratings score provides useful information about the perceived amount of Obama’s responsibility, it is not, by itself, fully informative in that it provides little sense of Obama’s responsibility relative to other political and non-political actors. Such insight requires the inspection of ratings scores for the other actors listed in Table 1.

OPEN IN VIEWER

Table 1.

Responsibility Ratings for Poor Economic Conditions in the United States.

	Obama	Bush	Democrats in Congress	Republicans in Congress	Wall Street bankers	Consumers	Lenders
A great deal	13.7	30.0	13.3	20.3	32.4	18.4	24.0
A lot	14.0	22.0	19.0	23.7	29.7	31.5	32.2
A moderate amount	22.7	25.5	36.2	34.5	24.8	31.8	21.4
A little	28.4	18.5	24.4	18.3	10.4	14.8	9.7
Not at all	21.2	4.1	7.0	3.3	2.6	3.5	2.6
Total %	100.0	100.0	100.0	100.0	100.0	100.0	100.0
Number of cases	5,832	5,834	5,750	5,749	5,706	5,824	5,801
M	0.43	0.64	0.52	0.60	0.70	0.62	0.72

Source. 2012 ANES.

Note. Table entries are the percentage of respondents who assigned the specified level of responsibility to each actor. Mean responsibility denotes the average level of responsibility attributed to each actor on a 5-point scale, with higher values denoting greater responsibility. Mean scores have been converted to a 0 to 1 range. ANES = American National Election Study.

A cursory inspection of the data in Table 1 quickly reveals two important insights on the nature of responsibility judgments in 2012. First, President Obama is rated as comparatively less responsible than each of the other six actors. His mean ratings score of 0.43 is considerably lower than those of the other three political actors: Bush (M = 0.64), Republicans in Congress (M = 0.60), and Democrats in Congress (M = 0.52). It is also lower than those of the three non-political actors: Wall Street bankers (M = 0.70), consumers who borrowed too much money (M = 0.62), and mortgage lenders (M = 0.72). Second, the results are consistent with Cutler’s (2008) observation that responsibility attributions are not a zero-sum game, as six of the seven actors receive a mean score above the midpoint of the ratings scale. Taken together, these data suggest that any effort to explain the electoral consequences of responsibility judgments must be careful to account for the competitive and comparative nature of the attribution process.

Constructing Comparative Responsibility Ratings

Using the data reported in Table 1, the proposed formula can be used to calculate a comparative responsibility rating for President Obama for each respondent. A maximum score is obtained from this formula when Obama receives the highest possible rating of 5 and each of the other six actors receives the lowest possible rating of 1. Such hypothetical ratings produce a comparative responsibility score of 9 [5 + (5 − (1 + 1 + 1 + 1 + 1 + 1) / 6) = 9]. A minimum score is obtained when Obama receives the lowest rating of 1 and all other actors receive the highest rating of 5, thereby producing a comparative score of −3 [1 + (1 − (5 + 5 + 5 + 5 + 5 + 5) / 6) = −3]. When converted to a 0 to 1 range, the comparative responsibility rating for Obama has a mean score of 0.41 (SD = 0.23).

Recall that the comparability portion of the formula is designed to capture the difference between the amount of responsibility attributed to the president and the mean amount of responsibility attributed to all other relevant actors. In its simplest computational form, the formula is implicitly based on a pair of key assumptions. First, during the formation of judgments about the comparative responsibility of a president, it assumes that all non-presidential actors are relevant and are weighted equally in the minds of citizens. Second, it assumes that responsibility ratings are somewhat discrete and that citizens do not group actors together on the basis of shared partisan ties.

Both assumptions can, of course, be questioned. Suppose, for example, that exercising electoral accountability mainly requires that voters judge the responsibility of the president relative to other political actors. If so, then any responsibility attributed to non-political actors such as bankers or consumers would be heavily discounted when calculating the comparative responsibility of the president. Given the competitive nature of responsibility judgments in politics, it is also possible that voters may group political actors together on the basis of shared partisan ties. Judgments about the comparative responsibility of President Obama, therefore, may be influenced not only by perceptions of Obama’s level of responsibility, but also by the level of responsibility ascribed to other Democratic actors such as congressional Democrats.

To explore these possibilities, we modify the process through which responsibility attributions are incorporated into the computational formula in two ways. First, we input only information about the perceived responsibility of political actors. ⁵ Second, we group actors together on the basis of shared partisan ties. These modifications produce a measure of comparative partisan responsibility that can be expressed as follows:

Comparative partisan responsibility = D + ( D − R ) .

In the present context, the D term denotes the mean level of responsibility attributed to President Obama and other Democratic actors (Democrats in Congress), whereas the R term denotes the mean level of responsibility attributed to Republican actors (President Bush, Republican actors). Under this formulation, comparative partisan responsibility is maximized when the president and members of his party receive the highest responsibility ratings and when representatives of the non-presidential party receive the lowest possible ratings. As before, scores range from a minimum of −3 to a maximum of 9. When converted to a 0 to 1 range, the comparative partisan responsibility index has a mean score of 0.44 (SD = 0.23).

It is not our contention that the logic underlying the measure of comparative partisan ratings is fundamentally different from the logic underlying the measure of comparative presidential ratings. Both are grounded in our belief that responsibility judgments are forged in a comparative evaluation process in which people judge the relative responsibility of competing actors. The measure of comparative partisan ratings merely relaxes the assumption that all actors compete independently in the minds of voters and proposes instead that voters may view some actors as competing together on the same team based upon shared partisan ties. The proposition that voters group actors together on the basis of shared partisan ties is well grounded in theories of responsible party government (Ranney, 1954; Schattschneider, 1942). ⁶ It is also consistent with empirical evidence showing that people often rely on motivated or partisan reasoning when making responsibility judgments (Nawara, 2015).

We now possess three different indicators of President Obama’s responsibility for the economy: a simple rating of presidential responsibility, a measure of comparative presidential responsibility, and a measure of comparative partisan responsibility. In the analyses to follow, we will examine the effects of each measure on three dependent variables of theoretical interest: presidential approval, presidential vote choice, and congressional vote choice. ⁷ Before doing so, however, we comment briefly on the construct validity of these measures by examining their relationships with known correlates of responsibility attributions.

A wealth of observational and experimental evidence has established that attributions of political responsibility are strongly tied to citizens’ partisanship and ideology (Gomez & Wilson, 2003; Hobolt et al., 2013; Malhotra & Kuo, 2008; Marsh & Tilley, 2010; Rudolph, 2003a, 2003b, 2006; Tilley & Hobolt, 2011). Simply put, people tend to engage in selective attribution such that they attribute responsibility for policy successes (failures) to their political allies (adversaries). A valid measure of responsibility attributions, therefore, should be strongly correlated with party and ideological identification. Similarly, responsibility attributions are politically consequential and exert a strong influence on both vote choice and political approval (Cutler, 2004; Feldman, 1982; Marsh & Tilley, 2010; Rudolph, 2003b; Rudolph & Grant, 2002). In the American context, a valid measure of responsibility attributions should be strongly correlated with presidential approval, presidential vote choice, and congressional vote choice.

Table 2 shows the bivariate relationships between the three measures of presidential responsibility and their known correlates. As expected, partisanship and ideology are positively correlated with conventional presidential ratings (r = .63, r = .50), confirming that Republicans and conservatives are more likely to blame President Obama for the nation’s poor economic conditions than are Democrats and liberals. Conventional presidential ratings are negatively correlated with presidential approval (r = −.72), presidential vote choice (r = −.74), and congressional vote choice (r = −.59). This implies that people who blame Obama are less likely to approve of Obama’s job performance, less likely to vote for Obama, and less likely to vote for a member of the president’s party.

OPEN IN VIEWER

Table 2.

Correlates of Responsibility Attributions by Type of Measurement.

	Party Id (GOP)	Ideology (CON)	Obama approval	Vote Obama	Vote Dem
Conventional presidential ratings
Obama responsibility	.63	.50	−.72	−.74	−.59
Comparative presidential ratings
Obama responsibility	.66	.52	−.74	−.76	−.62
Comparative partisan ratings
Obama/Democrats responsibility	.72	.56	−.77	−.78	−.65

Note. Table entries are bivariate correlations between the specified responsibility rating and the column variable. Party identification and ideology are coded with Republicans and conservatives at the high end of the scale. All correlations are statistically significant at the .01 level.

As can be seen in Table 2, the magnitude of all five correlations is somewhat larger when using comparative presidential ratings as the indicator of presidential responsibility instead of the conventional ratings. Within each of the five columns, however, the largest correlations are observed when using comparative partisan ratings as the indicator of presidential responsibility. ⁸ In terms of construct validity, the pattern of correlations in Table 2 suggests that both types of comparative ratings outperform conventional ratings as indicators of presidential responsibility. These findings underscore the importance of using measurement instruments that capture both the intensity and the comparability of responsibility judgments.

Overview of Hypotheses

In the next section, we exploit the analytical advantages of the panel component of the 2012 ANES to examine the performance of each measure of presidential responsibility in predicting citizens’ political evaluations. Each of the seven responsibility questions was administered during the pre-election wave of the survey, while two of the three dependent variables were measured in the post-election wave. ⁹ In the context of a sluggish economy, the extant literature creates clear expectations about the projected impact of citizens’ responsibility attributions on their political evaluations. Controlling for other factors, blaming President Obama for the nation’s poor economy is expected to decrease the likelihood of voting for Obama and of voting for a congressional Democrat. Blaming Obama is also expected to decrease presidential approval. Support for these three hypotheses is expected across all three measures of presidential responsibility. At issue in the analysis is whether one or both measures of comparative responsibility outperform conventional ratings as a predictor of political evaluations.

Multivariate Analysis

Table 3 reports the results of three ordered probit models of presidential approval in which response options include strongly disapprove, disapprove, approve, and strongly approve. The first data column reports the results when using conventional ratings as the measure of presidential responsibility. The second and third data columns report the results of similarly specified models when substituting comparative presidential ratings and comparative partisan ratings, respectively. Each model includes measures of respondents’ party identification, ideology, sex, race, and education as controls.

OPEN IN VIEWER

Table 3.

Determinants of Presidential Approval.

	Presidential approval	Presidential approval	Presidential approval
Presidential responsibility
Conventional ratings	−2.66** (0.09)	—	—
Comparative presidential ratings	—	−4.14** (0.12)	—
Comparative partisan ratings	—	—	−4.53** (0.16)
Control variables
Party identification (Republican)	−1.95** (0.08)	−1.83** (0.08)	−1.56** (0.09)
Ideology (conservative)	−0.85** (0.11)	−0.78** (0.10)	−0.64** (0.11)
Female	0.09* (0.04)	0.13** (0.04)	0.08* (0.04)
Black	0.80** (0.06)	0.80** (0.06)	0.76** (0.07)
Education	−0.25** (0.07)	−0.35** (0.07)	−0.21** (0.07)
τ1	−3.25 (0.08)	−3.76 (0.09)	−3.85 (0.09)
τ2	−2.54 (0.08)	−3.04 (0.08)	−3.14 (0.09)
τ3	−1.46 (0.07)	−1.95 (0.07)	−2.06 (0.08)
Number of cases	4,979	4,979	4,979
Log likelihood	−3,806.9	−5,331.6	−5,280.4
Pseudo R2	.41	.41	.41

Note. Table entries are ordered probit estimates with standard errors in parentheses. All independent variables have been transformed to share a common range of 0 to 1.

*

p < .05. **p < .01.

As can be seen in Table 3, the effects of the control variables comport well with theoretical expectations. The coefficients denoting party identification and ideology are both negatively signed and statistically significant across all three columns, confirming that Republicans and conservatives are less likely to approve of President Obama’s job performance than are Democrats and liberals. Presidential approval is higher among women, African Americans, and those with lower levels of education.

Of central interest in Table 3, though, are the effects of presidential responsibility for the economy. Consistent with expectations, the coefficient for the conventional ratings measure of presidential responsibility is negatively signed and statistically significant. As predicted, those who blame President Obama for the poor economy exhibit lower levels of presidential approval. When substituting the comparative presidential ratings or the comparative partisan ratings as the measure of presidential responsibility, the results are qualitatively similar. Attributions of presidential responsibility are consistently associated with lower levels of presidential approval regardless of which measure is used. Importantly, it should be noted that the size of the responsibility coefficients gets larger moving from left to right and that these differences are statistically significant. Parameter equality across columns was tested using Wald tests for non-linear models on non-independent samples as proposed by Clogg, Petkova, and Haritou (1995). ¹⁰

These tests demonstrate that the coefficient for comparative presidential ratings is larger than that for conventional ratings, χ²(1df) = 501.7, p < .01, and the coefficient for comparative partisan ratings is larger than those for both conventional ratings, χ²(1df) = 276.4, p < .01, and comparative presidential ratings, χ²(1df) = 13.2, p < .01. ¹¹

To better assess the magnitude of effects reported in Table 3, the top panel of Table 6 reports the predicted probabilities of presidential approval at selected levels of presidential responsibility. ¹² Consider first the effects of conventional ratings. A one-unit increase in the perceived level of Obama’s responsibility decreases the likelihood of presidential approval from 0.62 to 0.28, a decline of 34 percentage points. ¹³ The effects of comparative presidential ratings are slightly larger in magnitude. A similarly sized increase in the comparative presidential ratings index reduces the likelihood of approval by 36 percentage points. The largest substantive effects, though, are produced by the comparative partisan ratings. A one-unit increase in presidential responsibility on the comparative partisan ratings index decreases the likelihood of presidential approval by 40 percentage points.

Consider next the determinants of presidential vote choice. Table 4 reports the results of three similarly specified probit models of presidential vote choice, with only the measure of presidential responsibility varying across models. Across all three model specifications, Republicans and conservatives are less likely to vote for Obama than are Democrats and liberals. African Americans, by contrast, are more likely to vote for Obama. The coefficients for all three measures of presidential responsibility are negatively signed and statistically significant, although they again vary in magnitude. As hypothesized, people who blame President Obama for the nation’s economy are less likely to vote for him than for Mitt Romney. As was the case for presidential approval, the coefficients for the two comparative responsibility measures are larger than the coefficient for conventional ratings. Wald tests show that the coefficient for comparative presidential ratings is larger than that for conventional ratings, χ²(1df) = 188.0, p < .01, and the coefficient for comparative partisan ratings is larger than those for both conventional ratings, χ²(1df) = 107.7, p < .01, and comparative presidential ratings, χ²(1df) = 15.4, p < .01.

OPEN IN VIEWER

Table 4.

Determinants of Presidential Vote Choice.

	Presidential vote choice	Presidential vote choice	Presidential vote choice
Presidential responsibility
Conventional ratings	−2.78** (0.16)	—	—
Comparative presidential ratings	—	−4.58** (0.27)	—
Comparative partisan ratings	—	—	−5.58** (0.37)
Control variables
Party identification (Republican)	−2.81** (0.15)	−2.67** (0.15)	−2.34** (0.15)
Ideology (conservative)	−1.80** (0.24)	−1.78** (0.25)	−1.48** (0.25)
Female	−0.09 (0.08)	−0.05 (0.08)	−0.12 (0.08)
Black	1.35** (0.19)	1.33** (0.19)	1.24** (0.18)
Education	0.01 (0.14)	−0.11 (0.14)	0.01 (0.14)
Constant	3.63** (0.18)	4.27** (0.21)	4.61** (0.24)
Number of cases	3,365	3,365	3,365
Log likelihood	−647.7	−619.2	−610.0
Pseudo R2	.72	.73	.74

Note. Table entries are probit estimates with standard errors in parentheses. All independent variables have been transformed to share a common range of 0 to 1.

*

p < .05. **p < .01.

The substantive magnitude of the attribution effects is, once again, most easily observed through the presentation of predicted probabilities. The middle panel of Table 6 presents the predicted probabilities of presidential vote choice at the specified levels of presidential responsibility. A one-unit increase in the conventional ratings measure of presidential responsibility lowers the likelihood of voting for Obama from 0.72 to 0.37, a decline of 35 percentage points. As before, the effects are somewhat larger when using the comparative presidential ratings. A similarly sized increase in this measure reduces the likelihood of voting for Obama by 38 percentage points. The effects of presidential responsibility are substantially larger when using the measure of comparative partisan ratings. Here, the same one-unit shock reduces the likelihood of voting for Obama from 0.81 to 0.34, a drop of 47 percentage points. Collectively, this pattern of results demonstrates that the effects of comparative ratings on presidential vote choice are consistently larger than those of conventional ratings, with the comparative partisan ratings exerting the most influence.

Finally, we turn to the determinants of congressional vote choice in Table 5. Not surprisingly, Republicans and conservatives are less likely to vote for a congressional Democrat than are Democrats and liberals. The likelihood of voting for a congressional Democrat increases among African Americans and the well-educated. The coefficients representing attributions of presidential responsibility are negatively signed and statistically significant across all three columns, although the coefficients get progressively larger in magnitude from left to right. Substantively, this finding indicates that people who blame the president for the economy are less likely to vote for members of the president’s party in congressional elections. The coefficient sizes, though, again vary across columns. The coefficient for comparative presidential ratings is larger than that for conventional ratings, χ²(1df) = 66.9, p < .01, and the coefficient for comparative partisan ratings is larger than those for both conventional ratings, χ²(1df) = 51.6, p < .01, and comparative presidential ratings, χ²(1df) = 6.4, p < .02.

OPEN IN VIEWER

Table 5.

Determinants of Congressional Vote Choice.

	Congressional vote choice	Congressional vote choice	Congressional vote choice
Presidential responsibility
Conventional ratings	−1.09** (0.12)	—	—
Comparative presidential ratings	—	−1.72** (0.18)	—
Comparative partisan ratings	—	—	−2.03** (0.21)
Control variables
Party identification (Republican)	−2.09** (0.13)	−2.03** (0.13)	−1.88** (0.13)
Ideology (conservative)	−1.36** (0.18)	−1.31** (0.18)	−1.22** (0.18)
Female	−0.04 (0.06)	−0.03 (0.06)	−0.05 (0.06)
Black	0.27** (0.10)	0.26** (0.10)	0.22* (0.10)
Education	0.24* (0.11)	0.22* (0.12)	0.27* (0.12)
Constant	2.00** (0.12)	2.17** (0.13)	2.27** (0.13)
Number of cases	2,747	2,747	2,747
Log likelihood	−980.6	−975.6	−969.7
Pseudo R2	.48	.49	.49

Note. Table entries are probit estimates with standard errors in parentheses. All independent variables have been transformed to share a common range of 0 to 1.

*

p < .05. **p< .01.

To illustrate the magnitude of these effects, the bottom panel of Table 6 displays the predicted probabilities of congressional vote choice by measure of presidential responsibility. The pattern of results is quite similar to what has already been observed, although the magnitude of effects is somewhat smaller in two distinct ways. First, the effects of presidential responsibility on congressional vote choice are markedly smaller than they are on presidential approval and vote choice. This is likely because voters find it easier to make the connection between presidential responsibility and presidential evaluations than between presidential responsibility and congressional evaluations. Attributions of presidential responsibility do influence support for other members of the president’s party, but such coattail effects are smaller than they are for more direct evaluations of the president. Second, the differences in effect sizes across measures of presidential responsibility are somewhat smaller in the congressional context. When using the conventional ratings, a one-unit increase in presidential responsibility reduces the likelihood of voting for a member of the president’s party by 14 percentage points. When using the comparative presidential ratings, the same increase in presidential responsibility reduces that likelihood by 15 percentage points. The substantive effect of presidential responsibility climbs to 18 percentage points when using the comparative partisan ratings.

OPEN IN VIEWER

Table 6.

Effects of Responsibility Attributions by Type of Measurement.

	Low responsibility	High responsibility	First difference
Predicted probabilities of presidential approval
Conventional ratings
Obama responsibility	0.62 (0.02)	0.28 (0.01)	0.34
Comparative presidential ratings
Obama responsibility	0.63 (0.02)	0.27 (0.01)	0.36
Comparative partisan ratings
Obama/Democrats responsibility	0.68 (0.02)	0.28 (0.01)	0.40
Predicted probabilities of voting for Obama
Conventional ratings
Obama responsibility	0.72 (0.02)	0.37 (0.02)	0.35
Comparative presidential ratings
Obama responsibility	0.74 (0.02)	0.36 (0.02)	0.38
Comparative partisan ratings
Obama/Democrats responsibility	0.81 (0.02)	0.34 (0.02)	0.47
Predicted probabilities of voting for a congressional democrat
Conventional ratings
Obama responsibility	0.54 (0.02)	0.40 (0.02)	0.14
Comparative presidential ratings
Obama responsibility	0.54 (0.02)	0.39 (0.02)	0.15
Comparative partisan ratings
Obama/Democrats responsibility	0.57 (0.02)	0.39 (0.02)	0.18

Note. Table entries are the predicted probabilities of the specified dependent variable with standard errors in parentheses (sum of approval and strong approval in the top panel). For purposes of comparison, high/low responsibility is defined as ½ standard deviations above/below the mean, respectively.

Discussion

The assignment of political responsibility, we have argued, requires citizens to consider questions of both intensity and comparability. Few scholars, we suspect, would dispute the theoretical claim that intensity and comparability are key ingredients in the attribution process. The problem, however, is that our measures of responsibility attributions do not always do justice to our theories. Conventional measures of responsibility ratings emphasize the importance of intensity but neglect the importance of comparability. Rankings recognize the importance of comparability but are not designed to provide much information about intensity.

The current study addresses this problem and extends the attribution literature in three meaningful ways. First, from a conceptual standpoint, we propose a measurement strategy that can separately yet simultaneously incorporate information about intensity and comparability into a single attributional scale. Because it is constructed through the transformation of conventional ratings instruments, our measures of comparative responsibility ratings retain the administrative efficiency and cognitive simplicity of a classic ratings-based approach. By incorporating information about the comparative nature of attributional judgments, though, the comparative responsibility ratings also exploit the analytical advantages of a rankings-based approach. Second, from a methodological perspective, correlational evidence suggests that, as an indicator of presidential responsibility, both comparative presidential ratings and comparative partisan ratings have greater construct validity than do conventional ratings.

Finally, the results make a substantive contribution by demonstrating the consequences of citizens’ responsibility attributions during the 2012 election. Throughout the campaign, public attention was heavily focused on the nation’s poor economic conditions. A central issue in the election was the question of whether citizens would fault the economic policies of the Obama administration or whether they would blame former President Bush. Although President Obama was clearly judged to be less responsible than Bush, attributions of presidential responsibility proved to be quite consequential. People who blamed President Obama for the nation’s economy expressed much lower levels of approval and were much less likely to vote for the president or for members of the president’s party. It should be noted that this finding held across all three measures of presidential responsibility.

What, then, is the value added of using the comparative ratings approach to measuring presidential responsibility? At first blush, it may seem like a comparative ratings approach has little to offer as it generates essentially the same causal inferences as does a conventional ratings approach. Regardless of which measurement strategy is used, blaming the president for the nation’s poor economic performance was found to dampen levels of presidential support. Upon closer inspection, though, the choice of measurement strategy does appear to influence the magnitude of effects ascribed to citizens’ responsibility attributions. The observed effects of presidential responsibility are consistently larger when using comparative responsibility ratings, particularly the comparative partisan ratings. These findings suggest that, by ignoring the comparative nature of responsibility judgments, a conventional ratings approach may underestimate the full impact of citizens’ responsibility attributions on their political evaluations.

Recent work has contributed new insight about how the attribution process works outside of the economic domain, particularly in the area of foreign policy (Nawara, 2015; Sirin & Villalobos, 2011). Future research might fruitfully consider the utility of comparative responsibility measures in the foreign policy domain. Although we believe that the comparative responsibility formula can be applied to the foreign policy domain, we also believe that two considerations should be kept in mind when doing so. First, citizens tend to attribute more responsibility to the president vis-à-vis non-presidential actors in the foreign policy domain than in the domestic policy domain (Sirin & Villalobos, 2011). Although this would not affect the process through which the comparative responsibility formula is calculated, it could affect the values of the raw ratings that are incorporated into the formula. Second, the application of the comparative responsibility formula to the foreign policy domain may also have implications for how one decides to group actors together. If some potentially responsible actors are foreign or if a particular issue is not polarized along partisan lines, the argument for using the comparative partisan ratings may be somewhat weakened. If a foreign policy issue is polarized along partisan lines (e.g., the Iraq War), however, then using the comparative partisan measure may be entirely reasonable.