John Brehm and Paul Gronke
Duke University
Presented at the 1994 Annual Meeting of the American Political Science Association, New York City. We thank John Aldrich, Mike Alvarez, Regina Baker, Neal Beck, Paul Brace, Simon Jackman, and Dan Wood for comments, and Neal Beck for data.
Since Mueller (1973), the study of Presidential popularity routinely designates certain historical events as ``rallying'' events, especially the onset of foreign conflicts. Subsequent scholarship explores the effect of additional significant historical events (such as scandals or bad economic conditions) upon the President's stock of approval. This paper argues that prior research has misconceptualized ``rallies,'' which refer to a consolidation of consensus on the part of American attitudes, not just a boost in the President's average approval rating. The variance of approval is an important, but neglected, aspect of approval, and merits attention to the causes of systematic change in variance. The findings demonstrate, among other points, that Presidents gain more by performing the role of Commander-in-Chief than peacemaking statesman, and that Presidents can expect increasingly volatile approval ratings over the course of their administrations. These results have significant implications for the support of rational political actors in the legislature and for evidence of the rationality of public opinion.
* 144pt
Ample evidence demonstrates that approval ratings are a critical element of presidential leadership, especially in the legislature. When approval ratings are high, members of the president's party in Congress are less likely to be defeated in the midterm then when they are low. Presidents generally aim for a more ambitious agenda, and the president's legislative proposals are more successful, under conditions of high approval.1
What has been, heretofore, unexamined is the impact of volatility and consensus on presidential leadership. Rises and falls in mean levels of presidential approval - conventionally called rallies and busts - capture only one part of the picture; there are also changes in level of public agreement and disagreement about the performance of the president, or consensus, and the frequency and level of changes in approval, or volatility. Although the terms ``rallying'' and ``consensus building'' are de rigeur in speaking of the importance of events, the analysts proclaim these kinds of effects on the basis of shifts in the aggregate mean approval. From ordinary least squares through rarefied time-series, these are all approaches to modeling the movement in the mean. While ``rally'' clearly indicates a growth in approval, ``consensus'' has a very different meaning, one left unaddressed in the extensive literature on presidential approval. Consensus says nothing about the average; it does, however, imply that the variance in approval is relatively small. The principal insights of this paper are that volatility in presidential approval has increased over the postwar era, that there are regularized patterns of volatility during an administration, and that some categories of events lead to consensus in approval while others are associated with dissensus. This paper identifies the conditions under which presidential approval rises and falls (via its mean) and the variance of approval expands and contracts (via its variance).
There are real political reasons that the degree of consensus in approval ratings matters, along with its mean. Brace and Hinckley (1992) show how modern presidents are highly sensitive to changes in opinion polls. Presidents may delay certain actions because they cannot afford a drop in opinion polls, or may pursue other activities, especially foreign policy adventures, because they result in a boost in popularity. For example, George Bush received criticism for attempting to protect his 1991 Gulf War popularity surge into the 1992 election by purposely adopting a limited domestic agenda. Burbach (1995) discusses the risks presidents face when they attempt to create rallies by engaging in foreign policy adventures. A volatile public only increases these risks, because (under conditions of volatile approval) public opinion could turn away from the president as rapidly as it turned in favor.
Second, consensus itself is of intrinsic interest. As we show below, some foreign policy events have the unique feature of causing dissensus or disagreement among the public at the same time they boost approval. The extent to which the public is united behind a president's actions is directly measurable by the variance in approval. A president has a greater claim for popular legitimacy behind his actions when he can point to a consensual population, rather than to a population which is bitterly divided. Unlike the mean, which is a measure of the balance of support, consensus is a measure of the breadth of support.
Third, when looked at from the perspective of a rational political actor trying to decide whether to support or oppose the president, the possibility for future changes in approval introduces uncertainty. If job approval is relatively stable, Congress, the media, and the public have a consistent image of the president. Political actors can make calculations without bothering to include uncertainty. A president with a wildly variable base of support, in contrast, introduces a host of uncertainties into the political system. Should I tie my fortunes to a popular president today, when he may be unpopular come election day? Variance, consensus, and volatility matter.2 An important component of this research project requires that we attend to events. Extensive prior research (e.g., Mueller 1973, Brace and Hinckley 1992, Brody 1991) shows that significant political events have repercussions for the president's stock of popularity. Some events may be expected to boost approval: when enemy troops initiate conflict with the U.S. and its allies or when the president successfully completes a major domestic or foreign policy initiative. Other events erode the president's base of support: when the administration has to escalate involvement in an unpopular foreign conflict or when current or former members of the administration fall under indictment. But nearly all of the extant research on the effect of historical events treats the effect of events as one of two crude, positive and negative, categories.3 As a consequence, the analyst constrains the impact of events to two values. In other words, crude categorization of ``positive'' and ``negative'' events declares that warlike initiatives equate with peacemaking to boost the president's ratings, and that scandals, race riots, and strike-breaking equally undermine a president. One consequence of such crude categorization is that it misstates the importance of those events.4 A much more important flaw is that we lose information about the relative importance of widely dissimilar events across administrations. This research shows that not all events are alike, and that a much more nuanced understanding of presidential approval is easily obtained through more careful coding.
The benefit of our approach is that it allows us to understand what variables account for shifts in the presidential approval (the mean) as well as what variables account for changing magnitude of these shifts (the variance). The next section reviews the existing evidence as to why variance in approval may have increased over time.
``Do you approve or disapprove of the way President [name of incumbent] has handled his job as president?"Our dependent variable is the aggregate percentage of respondents who state that they approve of the president's handling of his job, or some transformation of this variable.5 This time series has formed the foundation for virtually every study of presidential popularity. The evidence for increased variability in these ratings is spotty and unsystematic. Edwards and Gallup report the standard deviations in presidential approval across presidencies, and conclude that ``instability has not increased steadily over time" (1990, p. 122). On the other hand, the low points of approval have decreased over the decades, and the range of approval ratings does seem to have increased, as shown in the final column of Table .
We argue that comparing standard deviations across administrations is the wrong way to proceed. The technique of accumulating the standard deviation of approval ratings compensates for truly unusual highs and lows, but does not recognize the turbulence of history. Compare administrations with a high frequency of events (e.g., the Johnson administration) with those with a low frequency of events (e.g., the Carter administration). The standard deviation of the two administrations are relatively close. Does the slightly higher standard deviation of approval under Johnson mean a more ``volatile'' public than that under Carter, even if fewer consequential events took place? Scholars of the presidency know that we cannot understand mean levels of presidential approval without knowing the particular circumstances of each administration, whether this includes economic performance, time in office, war casualties (Mueller 1973), media coverage of the administration (Brody 1991), or the ``dramatic events of a term'' (Brace and Hinckley 1992, p. 10). The evidence is convincing that any researcher modeling approval - the mean or the variance - must take into account specific events.
A cursory examination of the movement of approval, shown in Figure , highlights the centrality of events. The increases during Nixon's, Carter's, and Bush's terms are evident in both the table and the figure, as is the decline in variability during Reagan's term of office. Yet, even this figure by itself says nothing about different levels of variance in approval ratings across administrations. Was popular approval during Nixon's term particularly ``volatile,'' or are our eyes being misled by the Watergate plummet? Was Bush's approval rating really ``volatile'' (as Brace and Hinckley claim [1992, Ch. 7]), or was there simply a large surge and decline during and after the Gulf War conflict? No conclusive statements about variability can be made from these data, because we have no quick shorthand way to compare events across administrations. Certainly it does not make sense to compare the standard deviation during Reagan's term with the standard deviation across Nixon's term, for example, without accounting for the Vietnam War, Watergate, the assassination attempt, economic performance, and so forth. But how can we compare widely disparate administrations? We present an approach here that makes such a comparison possible, and allows us to isolate the existence, and causes, of variance in approval ratings.
Theory leads us to expect at least four reasons for different levels of consensus about the president at different times. First, we expect the president's stock of consensus will diminish over time. All presidents expect to benefit from a honeymoon period at the outset of the administration, where in- and out-partisans alike share good feelings about the president. This is a period of consensus. As the administration proceeds, presidential initiatives (or lack thereof) antagonize some interests and please others. At their outset, presidencies are blank slates. Few people have experiences with which they can judge performance, but many share high expectations. Each successive month adds to those experiences, and increasing numbers may render judgments positive and negative. The result will be that variance in approval increases over time.
Second, deliberation reduces variance. There is one obvious occasion when Americans are called to deliberate upon the performance of the president: election years. We know from prior research that pre-election polls are wildly variable, but decline in variance as the election comes closer. One interpretation of the reduction in variance is that the public is coming to a stable judgment about the candidates' probable performance (Gelman and King 1993). Re-elections are slightly different, in that there is a specific reference point in the form of the incumbent president; however, re-elections trigger partisan reactions to the incumbent president as well. The result should be that elections, particularly those involving lame duck presidents, will reduce variance in approval.
Third, events initiated by the president's administration should increase or decrease variance, depending on the nature of information the event provides to in- and out-partisans. Zaller's (1992) argument about differential information flows in the public contends that out-partisans are much more willing to accept negative information about the president than would in-partisans. (Equivalently, out-partisans resist information favorable to the president.) Different events should be expected to produce consensus, whether a consensus of failure (after revelations of scandal), or of success (after pronouncements of significant foreign or domestic policy accomplishments). Positive events lead to a reduction in variance, since they benefit both in- and out-partisans. Negative events, in contrast, should lead to an increase in variance. In-partisans are able to counter-argue while the out-partisans willingly adopt the new information. But it is possible that particularly strong negative signals, such as administration scandals, overcome hurdles to persuasion among in-partisans, and induce a reduction in variance.
In a related way, attacks by foreign states on the U.S. and its allies represent a different form of information, one which is rarely of an obvious partisan nature. We expect that the variance in approval should decline precipitously - consensus emerges - when the U.S. is attacked by another country. When the U.S. initiates conflict, however, domestic opposition is much more likely, especially in a ``pretty prudent'' post Vietnam public (Jentleson, 1992). Thus, we expect positive events, scandals, and foreign attacks on the U.S. will lead to declines in the variance in approval, while other negative events and U.S. initiated foreign conflict will lead to an increase in the variance.
Fourth, we expect a linear increase in variance across the postwar era as a consequence of declining partisan ties. Partisan affiliation and partisan attachments provide an inertial base in evaluations of the president. Voters who are of the same party as the president are more likely to resist negative information about him, and therefore less likely to incorporate such information into their approval of the president (Zaller 1992). Voters who are of the opposite party would be more likely to incorporate negative information, but less likely to incorporate positive information. Independent voters should be the most receptive to new information of either valence. These four hypotheses comprise the central goals of this research. Each implicates, to some degree, cross sectional and over time variance in presidential approval.
One problem for most researchers is that the pooled cross-sectional time series data are generally unavailable, whether due to cost or the cumbersome nature of the data itself. The aggregate time series, however, is easily obtained. Our technical innovation, outlined in the next section of the paper, allows us to recover analysis of the variance in cross-sections of approval ratings through an inferential approach, a modified ``autoregressive conditional heteroskedasticity'' (ARCH) model. The approach is tailored to test our hypotheses about changes and trends in volatility, and has widespread application to models where over-time variance is a dependent variable of interest.
|
|
|
Typically, the analyst chooses an appropriate function to reparameterize m in terms of the substantive explanatory variables, while leaving s2 as a constant:
|
There are, roughly speaking, two classes of models of heteroskedasticity in time-series. One approach (``multiplicative heteroskedasticity'') treats variance as a function of some set of explanatory variables, but is not conditional on variance in prior periods (Greene 1993, p. 405-407; Harvey 1990). An alternative approach (``autoregressive conditional heteroskedasticity'' or ARCH) treats variance as conditional on variance in previous periods, yet typically omits explanatory variables (Greene 1993, p. 438-442; Harvey 1990). We develop an estimator which combines the two.
The contrast between multiplicative heteroskedasticity and ARCH is straightforward. The multiplicative heteroskedasticity model represents variance as a function of some set of exogenous variables Z (which may or may not include the X variables in the function for the mean):
|
The ARCH approach, developed first by Engle (1982), represents variance as a function of the square of the residuals in the previous period (et-12):
|
|
As innovative and useful as the ARCH approach can be, in the present context it yields little of direct use. We gain more efficient estimates of the parameters on m, to be sure, but a standard ARCH approach doesn't provide information about what causes s2 to vary. Since we are interested in testing a hypothetical relationship between events, partisan strength, and variance in approval ratings, we need a different functional form, one which allows us to explain changes in s2. To this end, we develop a hybrid of the multiplicative heteroskedasticity and ARCH methods (referred to as ARCH-EDL). The autoregressive nature of the time series is retained, but we add a series of explanatory variables. The parameterization for the variance thus becomes:
|
The format of the exponentially distributed lag is straightforward. Let l represent the exponential decay and X represent a matrix of independent variables (lagged or contemporaneous):
|
The dependent variable is the mean level of presidential approval, taken from the Gallup series, and converted to monthly data.7 Our primary interest is in the model for the variance. Thus, while we need to rely upon a well specified model of the mean, we will not spend a great deal of space justifying our selection of a particular specification.8 There are four principal independent variables in the model: the change in unemployment (monthly first difference), rate of inflation, a dummy for Watergate, and the number of American troops killed in Vietnam (during Johnson's administration). In order to account for the delay before such information would be incorporated into approval ratings, all these independent variables are lagged one month.9 We add one of two sets of dummy variables to represent significant historical events in the model of the mean. The first set, following conventional practice, is taken from Brace and Hinckley (1991): one dummy denotes events favorable to the president, the other dummy denotes events which are unfavorable to the president.10 We do not lag the event variables, because we believe that the transmission of information about these events should have a very short term impact, unlike the effect of more complex information such as the change in unemployment or the level of inflation.
As already mentioned, we believe that this coding scheme is flawed. The conventional coding scheme developed incrementally, from Mueller's (1972) observation that presidential approval increases during ``rally'' events, to Kernell's (1986) fuller identification and analysis of rally events, to Brace and Hinckley's (1991) attempt to provide a relatively comprehensive list of ``positive'' and ``negative'' events. As Brody and Shapiro note, however, even supposed ``rally'' events (foreign policy adventures) differ dramatically in their impact on presidential approval (Brody 1991). By only selecting ``rally'' events, one precludes the ability to understand declines in presidential approval. Categorizing events solely into ``positive'' and ``negative'' is better, but still might obscure important differences among events.
A more informative approach, we believe, is that taken by Marra, Ostrom, and Simon (1990), where one includes separate dummy variables denoting different kinds of events.11 There are, broadly speaking, two areas of presidential activity, domestic and foreign, and two kinds of effects on approval, approval enhancing and approval diminishing. These categories reflect longstanding traditions of presidential policymaking (e.g. the ``two presidencies'') and approval research, and are not reviewed here. Since presidents are held particularly responsible for economic performance, we code economic news separately from other domestic policy events, and otherwise expect that positive economic news and domestic policy accomplishments will enhance a president's level of support, while poor economic news and policy failures will erode presidential support.
The other ``presidency'' is the foreign policy presidency. Scholars have long noted the president's greater flexibility and power in the foreign policy realm. Foreign policy events, however, are not unformly positive. Some, such as the Tet offensive or the U.S. downing of Libyan fighters, were associated with a decline in approval, while others, such as the Cuban missile crisis or the Vietnam peace agreement, show the commonly expected pattern, a short term boost for the president (Brody, 1991, Table 3.1). We distinguish between peace-making initiatives and those that involve the use of troops, and between conflicts begun by the U.S. and those started by enemy nations. It is entirely plausible - indeed, we will demonstrate as much - that the effect of enemy attacks is especially strong in boosting presidential approval, whereas conflicts initiated by the U.S. might find doubters among the respondents as to the wisdom of the policy.
Finally, presidents are judged on personal characteristics: competence, integrity, empathy, and leadership (Hinckley, 1990; Kinder, et al., 1980). While few events can be said to reflect directly on presidential character, the public responds to events that seem to bear on presidential fitness, character, and ability to lead. Thus, we anticipate a rally when the president's health is threatened and after an assasination attempt; and expect a decline in approval during administration scandals. Because of the extraordinary set of events surrounding Watergate, we include a separate dummy variable for time points compirising this scandal.12 These expectations are reflected in the coding of events shown in the table below. (A full listing of events and a more extensive description of the coding rules appears in Appendix A.) We expect that two of these should have negative signs in the model for the mean: adverse economic events and administration scandals. We expect that four of these should have positive effects upon the mean: domestic and foreign policy accomplishments, and U.S.- and enemy-initiated foreign conflicts. We are agnostic as to the effect of escalation.
Second, we expected that elections, as moments of deliberation, should decrease variance in approval. Hence, we include a dummy variable, election years, and expect that the underlying variance in election years should be considerably more narrow than it is in non-election years. Since we expect that the effect should be slightly less pronounced during re-election years, we also include an appropriate dummy measure.
Third, due to the differential effect of the events of an administration upon in- and out-partisans, we include one of two sets of dummy event codes into the variance portion of the equation, thus controlling for the unique history of each administration. This allows us to test whether particular kinds of events are more likely to lead to consensus while other events lead to dissensus. One set of events is Brace and Hinckley's (1992) positive and negative event codes; the other is our more elaborated set of events specific to different categories.
Finally, there are good grounds to expect that declining partisan ties in the postwar era may have undercut the president's ability to maintain a stable level of support, thereby leading to an increase in variance. We include three surrogates for strength of partisan attachment, the percentage of independent identifiers, as measured by the Gallup survey, the fraction of party defectors in each presidential election, and the number of split ticket voters (President-House) in each election.13 If our hypothesis holds, some or all of the proxy measures for declining partisan attachment should be positively signed in the variance model.
Adding events to the model of the mean yields substantial improvement in fit over Beck's specification; the difference in log-likelihood is significant at the p<.001 level (a c2 of 45.6). All else being held equal, a positive event is associated with an 3.55 point improvement in the president's approval rating, whereas a negative event leads to a 1.25 decline in approval. By these estimates, positive events boost a president by nearly twice as much as negative events undercut the president's approval. (The overall goodness of fit of the two models is beyond p<.001. Note that the two goodness of fit measures are not comparable, because the more elaborate events model is not nested within the simpler model.)
The advantage of the more elaborate coding of events is readily apparent: not all ``positive'' or ``negative'' events accomplish as much for the president's approval rating. First, consider the effect of significant economic events: presidents are penalized by 2.7 points for adverse economic news, over twice the effect of a generic ``negative'' event by the original coding.15 News of administration scandals did not undermine presidential popularity to a significant degree, a finding quite contrary to conventional accounts. Although the estimate is of the anticipated sign (negative), it is substantively trivial. We do note that the presence of the Watergate dummy variable removes the effect of the extraordinary sequence of scandalous events during Nixon's administration (when there were over fifteen events coded as administration scandals from February 1973 through Nixon's resignation in August 1974). This result stands in contrast to the findings of Ostrom and Simon (1985) where administration scandals, undifferentiated from Watergate, were the single most damaging category of events for a president. After controlling for the unusual scandals of Watergate, we found no significant effects of scandal on the president's approval.
Presidents receive greater credit for domestic accomplishments than foreign accomplishments, running somewhat against the received wisdom that presidents should focus on foreign policy since that is where they receive the most credit (although see Burbach, 1995). A domestic policy accomplishment, on average, produces a four point gain in the president's approval rating. Ostrom and Simon (1985) found that domestic policy initiatives (distinguished from completion of policy goals) were potentially a debit to the president (although that effect was not significant). Our results demonstrate that when presidents complete their policy goals, they reap significant popular rewards. On the other hand, a foreign policy accomplishment yields less than half that of a domestic policy accomplishment (and it is not statistically significant).
We differentiated between US-initiated and enemy-initiated foreign conflicts because we expected that it was possible that the events would not be symmetric. As Brody and Shapiro (Brody 1991, Ch. 3) and Jentleson (1992) note, public opinion about foreign policy events differs in meaningful ways. The storied ``rally-around-the-flag'' would be more likely in the presence of a foreign threat, but less likely when the U.S. entwines the military in an unprovoked engagement. Our suspicions were supported by the data: an enemy-initiated conflict was the most substantively significant category of events, accounting for over a four point increase in presidential approval (in excess of the expected effect of a generic ``positive'' event). US-initiated conflicts returned two additional points on the approval ratings, but these were not statistically significant. Likewise, escalation of continuing U.S. military involvement undercut presidential approval, but not by much, and not to a significant degree. Another angle on interpreting the relative importance of these events is to note that a president gains most when an enemy initiates conflict, more than twice the gain from a peaceful foreign policy accomplishment: the president gains more as commander-in-chief than he does as chief statesman.
Second, our hypotheses about election years and time in office are confirmed. Election years tighten the variance significantly (the coefficient estimate is statistically significant at p<.01 in a one-tailed test), indicating that, during these years, the population moves toward greater consensus in their evaluations of the president. This effect is not as large in re-election years ( -1.79 + .85 = -.94). This is not surprising. Re-elections activate all manner of partisan attitudes, requiring judgment of the president's performance. Those years which are both election years and are not re-elections are years in which the ``lame-duck'' president is completing the last months of his administrations.
In an opposite direction, the public becomes increasingly dissensual in assessments of the president's performance the longer the president has been in office. The percent of the president's term that he has completed tends to increase variance. The result is consistent with an argument that activist presidents alienate some groups while appeasing others, tending to increase variance over time. Presidents not only experience an inevitable decline in popularity during their term (Mueller 1973; Stimson 1976), but they also encounter increasingly conflicted opinion.16 This is an important result because it implies that presidents have the greatest consent behind their program in the earliest months of their administrations. In most ways, this elaborates the general honeymoon effect, with the difference here being that presidents can expect not just high approval, but uniformly high approval at the outset of their administratiosn.
What of the effect of events on variance? The importance of discriminating among events is made clear when one compares columns one and three in Table . When we estimate our model using the positive/negative coding scheme (column one), we find no statistically discernible impact of events on variance. The estimates from the first model (employing positive/negative coding) do provide one insight: positive events, which boost the mean level of approval, also increase the variance (by a predicted .77 points); whereas negative events, which eat away at approval, decrease variance by .44 points. Although statistically insignificant, the coefficient estimates indicate that positive and negative events have very different impacts on the variance of presidential approval as well as on the level of presidential approval. Does our coding scheme provide additional insights?
We believe that it does. One noticeable difference is that the second model produces statistically discernible coefficients on some events. We are much more interested, however, in the distinctive patterns among events. Our results show that public approval of the president becomes more consensual during domestic events. Each of the following variance coefficients is negative (i.e. associated with a decline in variance): scandals, adverse economic events, and domestic accomplishments (the one non-domestic category is ``foreign accomplishments''). The effect of domestic accomplishments on unifying public opinion seems especially strong, outweighing even the consensus that builds during an election year. Foreign events, in contrast, cause dissensus; U.S. led escalation in an ongoing conflict, U.S. initiated conflicts, and enemy initiated conflicts (contrary to our expectations) all have a positive impact in the variance side of the model. Not surprisingly, enemy attacks on the U.S. have the largest impact, also rivaling the election year effect. With the exception of foreign accomplishments, these results can be grouped into two general categories: war and peace. Wars, as shown in the model of the mean, produce a generalized boost for the president, but at the same time produce increased variance in public approval. It is hard to characterize the event as a ``rally'' if it does not also imply narrowing variance in approval.17 Events unrelated to war (all of them domestic affairs) have a mix of effects on the mean level of approval, depending on whether they are good (accomplishments) or bad (scandals, adverse economic events), yet on average are associated with a reduction in variance. These results are not inconsistent with the results using Brace and Hinckley's codes, when one realizes that the bulk of ``positive'' events are foreign affairs, ``rally'' events, whereas the bulk of ``negative'' events consist of scandals and bad economic news. Still, the two point coding disguises an important difference in public reactions to foreign and domestic affairs, a distinction which has characterized public opinion research for decades (see Holsti 1992 for a review). Our research supports the claim that public opinion on foreign policy can be just as influential as on domestic issues - as long as we pay close attention to different kinds of foreign policy events (Aldrich, et al. 1989). Furthermore, the distinction that we make among foreign policy events confirms the findings of Page and Shapiro (1992, p. 281): ``the American public makes many clear and reasonable distinctions among alternative [foreign] policies'' (emphasis added). Equally, these results confirm an argument of Brace and Hinckley (1992, p. 107-114) that the public reacts in very different ways to presidential use of force, depending on the perception of U.S. interests. These important differences - between public opinion on domestic and foreign policy and among kinds of foreign policy actions - are fundamental to our understanding of presidential approval.
There is only weak support for the theoretical expectation of positive signs on the effects of mixed partisan events and negative signs on the partisan-neutral events. Only two events achieve statistical significance here, and they are quite different: domestic accomplishments and administrative scandals. At a stretch one might regard these events as those which exceed hurdles to persuasion in a partisan environment, but there is no support for such a claim in the literature. Failure to confirm the effects of mixed partisan events with aggregate data does not mandate rejection of the hypothesis, but does argue for tests at a less aggregated level such as the pooled cross-sections.
The hypothesis about the impact of declining partisan ties, on the other hand, is not confirmed in any way. No matter how we measure it, we find no statistically discernible effect of withering partisanship on the variance of presidential approval. None of the following have statistically or substantively meaningful coefficients: the percent of partisans that defect, the percent of split ticket voters, or the percent of independent identifiers. Even though one might have expected declining partisanship to accommodate a more volatile public, the causes of variance lie elsewhere. Historical events and regular structural features of the presidency (known terms and regular elections) are much more important.
Finally, the ARCH-EDL model allows us to capture patterns and periods of changes in variance. An illustration of the predicted variance ([^(s)]), plotted in Figure , reveals a fairly regular pattern to variance across administrations, from a starting value of 3-3.5%, increasing to 4-4.5% by the end of a term. Compare this level of error to the sampling error for the Gallup series during most of this period, about 3%. In other words, for most of the series, the variance (at the one standard deviation level) is already larger than the sampling error (at the two standard deviation level). Were one to increase the variance to a comparable two standard deviation (95% confidence level), it is quite apparent that sampling error accounts for only a small proportion of the underlying variance in approval. Note also that there are only two administrations which break with the general pattern of rising variance: Ford (which was perhaps too short to identify the pattern), and Bush. Even Nixon's administration, with the exception of the final year, follows more or less a steadily rising pattern.
One can also readily see the estimated periods of low variance during election years. Every election year with the exception of 1968 and 1972 is marked by an unusually low estimated variance in approval. The former was an exceptionally eventful year, with not only significant events in Vietnam and at home, but also the Soviet invasion of Czechoslovakia. The latter involved both an announcement of a peace accord, and an escalation of bombing in Cambodia. There are multiple sharp spikes in estimated variance, due to a mix of both positive and negative events. The turbulence of Watergate is obvious, as was also the emerging consensus about Nixon's performance just prior to his resignation in August 1974.
Substantively, we found that variance increased only slightly over the forty years of the presidential approval series. This finding puts our work in concurrence with Edwards and Gallup (1990), but on much stronger methodological foundations since we control for the different events. The very slight increase in underlying variance in approval also tends to rule out the importance of such strong secular trends as declining partisanship.
Instead, we find that variance tends to increase over the course of each administration. Only two administrations buck the pattern of steadily rising variance: Ford and Bush (this latter finding places us in sharp contrast to Brace and Hinckley 1992, Ch. 7). And the Ford administration may have been too short to see such patterns, coming quickly into an election year. The underlying variance declines during election years, but especially during those years when a lame-duck is in office.
We find that history matters, and significant events can wildly disrupt the steady patterns of increasing variance on the whole, punctuated by lower periods during elections. With respect to the president's average level of approval, we find that historical events can be more important than economic conditions. Specifically, the president gains most by his role as commander-in-chief, in that his approval ratings climb more when the U.S. becomes involved in a foreign conflict. This is especially so when an enemy nation initiates the conflict (where the effect is twice the boost that the president receives when the U.S. initiates conflict).
Domestic politics matter, too. Bad economic news is the worst of the seven categories of historical events to befall a president (indeed, the only ``negative'' event that is statistically significant). A president also gains significantly by finalizing domestic policy, to a point nearly equal with that of the boost a president receives from enemy-initiated conflicts.
These results have a variety of implications for strategic political actors and for the rationality of the American public. Consider the effect of steadily increasing variance in approval upon the president's support in the legislature. We know that strategic political actors consider the president's level of popular support when deciding whether to support the president's program (e.g., Rivers and Rose 1985, Ostrom and Simon 1985). One implication of our work arises if strategic actors conduct a kind of hypothesis test on the president's approval levels. As the variance in presidential approval increases, this means that the strategic actor should begin to discount modest approval levels. Consider two presidents with approval levels at 60 percent, but one near the start of his administration (and low variance), and the second further along in the administration (with higher variance). The strategic actor should consider that the approval levels for the former need not slip very much in the short run, but that approval levels for the latter might slip significantly below 50 percent. In other words, it is not just the president's declining level of approval, but wider variance in approval which undermines the president's program.
There are also speculative implications to be drawn from our work with respect to the general rationality of the public. The alleged indifference of the American public to foreign affairs (Almond 1950) obviously does not hold with respect to three of the significant categories of events. Foreign conflicts, whether U.S.- or enemy-initiated, and foreign policy accomplishments have significant effects upon the president's approval ratings. Not only does public attention to these events tend to boost the president's base of support, it does so to a degree greater than most of the domestic events we study. Recent scholarship supports the argument that there are significant aspects of foreign policy events that affect attitudes towards political figures (e.g., Aldrich, Sullivan, and Borgida 1989), and in attitudes towards policy choices (e.g., Page and Shapiro 1992; Holsti 1992). Our work not only adds to these more general findings, but provisionally demonstrates that these events need not be ``rallying'' in the sense that they produce underlying consensus.
Lastly, the systematic pattern of declining variance in approval during election years deserves interpretation. Plainly, the greatest declines in variance in approval happen when the president is of limited relevance as a political actor: during the election year of the lame duck president's final year in office. At this time, one may reasonably expect that much of the partisan bickering over the actor would turn towards evaluation of the current political candidates. It is perfectly reasonable to expect that attitudes toward the lame duck president would stabilize as political attention drifts elsewhere.
Presidential approval has held the interest of political observers for almost fifty years. The typical conception of approval, which focusses on shifts in the mean, is incomplete. No one would argue that a statistical distribution can be described solely by its central tendency; nor should our exploration of the patterns and causes of presidential approval look only at the mean. As important in many circumstances is the level of agreement within the population - consensus - and the frequency and sharpness of shifts in public sentiment - volatility. In this paper, we derive an estimator for variance in a time series that can be easily incorporated into existing models of approval, and can be readily adapted to test hypotheses about the causes of changing volatility. We test three such hypotheses here: whether volatility displays trends over time and through administrations (it does); whether there is a differential impact of events, some causing public agreement and others disagreement (there are); and whether changes in the partisan makeup of the population has led to increasingly volatile approval (it has not). Still, whether due to changes in the nature of political information, changes in the occupants of the Oval Office, or simply a consequence of the tumult of the past forty years, presidents have had to contend with an increasingly fickle public.
Notes
Aldrich, John H., John L. Sullivan, and Eugene Borgida. 1989. ``Foreign Affairs and Issue Voting: Do Presidential Candidates `Waltz Before a Blind Audience?' `` American Political Science Review 83(March):123-142.
Almond, Gabriel. 1950. The American People and Foreign Policy. New York: Harcourt, Brace.
Beck, Nathaniel. 1991. ``Comparing Dynamic Specifications: The Case of Presidential Approval.'' Political Analysis 3: 51-88.
. 1989. ``Estimating Dynamic Models Using Kalman Filtering.'' Political Analysis 1: 121-156.
Bollerslev, Tim. 1992. ``ARCH Modeling in Finance: A Review of Theory and Empirical Evidence.'' Journal of Econometrics 52(1-2): 55-60.
Brace, Paul and Barbara Hinckley. 1992. Follow the Leader: Opinion Polls and Modern Presidencies. New York: Basic Books.
. 1991. ``The Structure of Presidential Approval." Journal of Politics 53(November): 993-1017.
Brody, Richard A. 1991. Assessing the President. Stanford, CA: Stanford University Press.
Burbach, David T. 1995. ``The Rational Public and the Rally Effect.'' Paper presented at the 1995 Annual Meeting of the American Political Science Association, Chicago, IL.
Edwards, George C. III and Alec M. Gallup. 1990. Presidential Approval: A Sourcebook. Baltimore, MD: The Johns Hopkins University Press.
Engle, R. 1982. ``Autoregressive Conditional Heteroskedasticity with Estimates of the Variance of United Kingdom Inflations.'' Econometrica 50(July): 987-1008.
Erikson, Robert S., Norman R. Luttbeg, and Kent L. Tedin. 1991. American Public Opinion: Its Origins, Content, and Impact, Fourth Edition. New York: Macmillan Publishing Company.
Gallup, George. 1984-1993. The Gallup Poll. Wilmington, DE: Scholarly Books.
Gelman, Andrew and Gary King. 1993. ``Why Are American Presidential Election Campaign Polls So Variable When Votes Are So Predictable?'' British Journal of Political Science 23(October): 409-452.
Harvey, A. 1990. The Econometric Analysis of Time Series, 2nd Edition. Cambridge, MA: MIT Press.
Hinckley, Barbara. 1990. The Symbolic Presidency. New York, NY: Routledge.
Holsti, Ole R. 1992. ``Public Opinion and Foreign Policy: Challenges to the Almond-Lippmann Consensus. Mershon Series: Research Programs and Debates.'' International Studies Quarterly 36(December): 439-66.
Jentleson, Bruce W. 1992. ``The Pretty Prudent Public: Post Post-Vietnam Amerian Opinion on the Use of Military Force.'' International Studies Quarterly 36 (March): 49-74.
Kernell, Samuel. 1986. Going Public. Washington, D.C.: Congressional Quarterly Press.
. 1978. ``Explaining Presidential Popularity.'' American Political Science Review 72(June): 506-522.
Kinder, Donald R., M.D. Peters, R.P. Abelson, and S.T. Fiske. 1980. ``Presidential Prototypes.'' Political Behavior 2(4):315-338.
Kinder, Donald R. and David O. Sears. 1985. ``Public Opinion and Political Action,'' in Lindzey and Aronson, eds., Handbook of Social Psychology. New York: Random House.
King, Gary. 1993. ``The Methodology of Presidential Research.'' In Edwards, Kessel, and Rockman (eds), Researching the Presidency. Pittsburgh, PA: University of Pittsburgh Press.
King, Gary. 1989. Unifying Political Methodology. New York: Cambridge University Press.
King, Gary and Lyn Ragsdale. 1988. The Elusive Executive. Washington, DC: Congressional Quarterly Press.
Krosnick, Jon A. and Matthew K. Berent. 1993. ``Comparisons of Party Identification and Policy Preferences: The Impact of Survey Question Format,'' American Journal of Political Science, 37(August): 941-964.
MacKuen, Michael B., Robert S. Erikson, and James A. Stimson. 1989. ``Macropartisanship." American Political Science Review, 83(December): 1125-1143.
Marra, Robin, Charles Ostrom, and Dennis Simon. 1990. ``Foreign Policy and Presidential Popularity.'' Journal of Conflict Resolution 34(December): 588-623.
Mueller, John. 1973. War, Presidents, and Public Opinion. New York: John Wiley.
Neustadt, R.E. 1980. Presidential Power: The Politics of Leadership. New York: John Wiley.
Ostrom, Charles M. and Dennis M. Simon. 1989. ``The Man in the Teflon Suit?'' Public Opinion Quarterly 53(Fall): 353-387.
---. 1985. ``Promise and Performance: A Dynamic Model of Presidential Popularity." American Political Science Review 79(June): 334-358.
Page, Benjamin I. and Robert Y. Shapiro. 1992. The Rational Public: Fifty Years of Trends in American's Policy Preferences. Chicago: University of Chicago Press.
Rivers, Douglas, and Nancy L. Rose. 1985. ``Passing the president's program: Public opinion and presidential influence in Congress.'' American Journal of Political Science 29(May):183-96.
Simon, Dennis M. and Charles M. Ostrom. 1989. ``The Impact of Speeches and Travel on Presidential Approval." Public Opinion Quarterly 53(Spring): 58-82.
Stanley, Harold W., and Richard G. Niemi. 1995. Vital Statistics on American Politics. Washington, D.C.: CQ Press.
Stimson, James. 1976. ``Public Support for American Presidents: A Cyclical Model.'' Public Opinion Quarterly 40(Spring): 1-21.
Zaller, John. 1992. The Nature and Origins of Mass Opinion. New York: Cambridge University Press.
| President | High Point | Low Point | Difference | SD |
| Eisenhower | 79 | 48 | 28 | 6.9 |
| Kennedy | 83 | 56 | 28 | - |
| Johnson | 79 | 35 | 44 | 13.5 |
| Nixon | 67 | 24 | 43 | - |
| Ford | 71 | 37 | 34 | 12.2 |
| Carter | 75 | 28 | 47 | 12.2 |
| Reagan | 68 | 35 | 33 | 7.7 |
| Bush | 83 | 32 | 51 | 15.1 |
| Type of Event | Approval | Approval |
| Enhancing | Diminishing | |
| Domestic | Positive Economic News | Negative Economic News |
| Policy Accomplishment | Policy Failure | |
| Foreign | Foreign Attacks | U.S. Attacks |
| Policy Accomplishment | Policy Failure | |
| Personal | Health | Administration Scandal |
| Assassination Attempt |
| Variable | BH Event | Std Err | New Event | Std Err |
| Mean Model | ||||
| Constant | 5.15 | 1.05 | 4.51 | 1.05 |
| Inflationt-1 | -0.15 | 0.05 | -0.12 | 0.05 |
| Unemploymentt-1 | -0.50 | 0.87 | -0.26 | 0.81 |
| Watergatet-1 | -1.42 | 0.89 | -1.64 | 1.09 |
| Vietnamt-1 | -1.70 | 0.58 | -1.18 | 0.58 |
| Positive Eventt | 3.55 | 0.76 | ||
| Negative Eventt | -1.25 | 0.51 | ||
| Adverse Economic Eventt | -2.70 | 1.26 | ||
| Administration Scandalt | 0.09 | 0.43 | ||
| Domestic Accomplishmentt | 3.76 | 1.10 | ||
| Foreign Accomplishmentt | 1.55 | 1.03 | ||
| Escalationt | -0.33 | 1.78 | ||
| US-initiated Foreign Conflictt | 2.16 | 1.36 | ||
| Enemy-initated Foreign Conflictt | 4.15 | 1.34 | ||
| Approvalt-1 (l) | 0.92 | 0.02 | 0.92 | 0.02 |
| Moving Average (q) | -0.18 | 0.06 | -0.17 | 0.07 |
| Variance Model | ||||
| a0 | 2.47 | 0.90 | 2.89 | 1.00 |
| a1 | 0.02 | 0.01 | 0.03 | 0.01 |
| Re-election Yeart | 0.57 | 0.53 | 0.85 | 0.48 |
| Election Yeart | -1.44 | 0.48 | -1.79 | 0.42 |
| Positive Eventt | 0.77 | 0.60 | ||
| Negative Eventt | -0.44 | 0.40 | ||
| Adverse Economic Eventt | -0.79 | 0.76 | ||
| Administration Scandalt | -1.02 | 0.19 | ||
| Domestic Accomplishmentt | -2.32 | 0.94 | ||
| Foreign Accomplishmentt | -0.53 | 0.74 | ||
| Escalationt | 0.30 | 1.34 | ||
| US-initiated Foreign Conflictt | 0.14 | 1.27 | ||
| Enemy-initated Foreign Conflictt | 1.64 | 1.04 | ||
| Percent Term Completedt | 0.97 | 0.61 | 1.62 | 0.57 |
| % Partisans Defecting | ||||
| in Presidential Votet | 0.03 | 0.04 | 0.04 | 0.05 |
| % Split Votes (Pres.-House)t | 0.00 | 0.06 | 0.03 | 0.06 |
| % Independentst | 0.05 | 0.08 | 0.03 | 0.07 |
| -2 ×(log\cal L0 - log\cal L) | 70.6 | 72.8 | ||
(continued)
Date Event BH BG
03/1949 Coal strike N NE
10/1949 Steel strike N NE
07/1950 North Korea attacks South Korea P EN
09/1950 Wage-price controls N NE
09/1951 Truman fires MacArthur N O
04/1952 Japanese Peace Treaty announced P FP
04/1952 Truman nationalizes steel industry N NE
06/1952 Court rules against Truman on steel case N O
03/1953 Soviets fire on U.S. bomber P EN
08/1953 Korean armistice announced P FP
10/1953 Eisenhower invokes Taft-Hartley N NE
07/1955 Soviets shoot down U.S. spy plane P EN
10/1955 Eisenhower has heart attack P H
06/1956 Eisenhower has major surgery P H
10/1957 Eisenhower orders army to Little Rock N PR
10/1957 Sputnik launched N O
06/1958 Sherman Adams scandal breaks N AS
07/1958 Eisenhower sends Marines to Lebanon P US
07/1959 Steel strike N NE
11/1959 Eisenhower invokes Taft-Hartley N NE
05/1960 U-2 incident P US
05/1961 Bay of Pigs invasion P US
08/1961 Berlin Wall crisis P EN
11/1961 Second Berlin Wall crisis P 3-?
03/1962 First American orbits Earth P DP
05/1962 Steel crisis N NE
10/1962 Integration crisis in Mississippi N PR
11/1962 Cuban Missile Crisis P EN
05/1963 Integration crisis in Alabama N PR
05/1965 Dominican Republic crisis P ES
08/1965 Vietnam draft doubled N ES
04/1966 Vietnam protests N PR
08/1966 Race riots in Chicago N PR
09/1966 Race violence in Atlanta N PR
08/1967 Race riots N PR
11/1967 Vietnam protest N PR
02/1968 Tet offensive N EN
04/1968 Johnson announces end to bombing P FP
05/1968 Campus protests N PR
09/1968 Soviets move into Czechoslovakia P EN
11/1968 Johnson halts bombing in Vietnam P FP
12/1968 Lowest unemployment in Fifteen years P PE
04/1969 Campus protests about Vietnam N PR
08/1969 Successful moon launch P DP
12/1969 Huge antiwar rally N PR
06/1970 Cambodia invasion N ES
06/1970 Protest killings at Kent State N PR
02/1971 Laos invasion N ES
04/1971 Antiwar demonstrations N PR
09/1971 Nixon imposes wage-price controls N NE
02/1972 Vietnam peace proposal announced P FP
04/1972 Increase in war and bombing N ES
01/1973 Vietnam peace accord P FP
02/1973 Watergate burglars convicted N AS
03/1973 McCord letter Sirica N AS
05/1973 Ervin Committee begins N AS
06/1973 Price freeze announced N NE
07/1973 Dean testifies N AS
08/1973 Agnew investigation revealed N AS
09/1973 Ehrlichman, Liddy, and others indicted N AS
10/1973 Saturday night massacre N AS
11/1973 Gap in tape revealed N AS
11/1973 Six Watergate figures sentenced N AS
04/1974 House judiciary hearings begin N AS
04/1974 Nixon ordered to pay back taxes N AS
05/1974 Judiciary hearings continue N AS
08/1974 U.S. v. Nixon announced (8/30) N AS
08/1974 Articles of Impeachment voted N AS
08/1974 Tapes incriminate Nixon N AS
10/1974 Ford pardons Nixon N O
05/1975 Cambodia falls N EN
06/1975 Mayaguez incident P EN
09/1978 Camp David Accords signed P FP
12/1979 Hostages first seized in Iran P EN
01/1980 Soviets invade Afghanistan P EN
02/1980 Inflation sets new record high N NE
04/1980 Helicopter rescue plan fails N US
05/1980 Race rioting N PR
03/1981 Assassination attempt on Reagan P H
08/1983 Soviets attack Korean airliner P EN
10/1983 Grenada invasion P US
03/1984 Record deficit balance of payments N NE
04/1984 Bombing of Nicaraguan harbors divulged N US
01/1985 Cabinet shakeup N O
04/1985 Bitburg controversy N O
07/1985 Hostage incident (06/14-06/30) P EN
08/1985 Reagan Surgery (07/13) P H
01/1986 Space shuttle explodes P O
04/1986 Libyan hostilities P EN
05/1986 Air strike on Libya P US
11/1986 First Iran-Contra revelation N AS
12/1986 Reagan claims Iran-Contra ignorance N AS
03/1987 Tower Committee report N AS
03/1987 Donald Reagan resigns N AS
05/1987 Iran-Contra hearings N AS
05/1987 Persian Gulf attack on U.S. P EN
06/1987 Iran-Contra hearings continue N AS
06/1987 U.S. escorts Kuwaiti tankers P US
07/1987 Iran-Contra hearings continue N AS
10/1987 Stock market plunges N NE
11/1987 Iran-Contra report by Congress N AS
12/1987 U.S.-U.S.S.R. treaty signed P FP
01/1988 Meese investigation N AS
04/1988 Justice Department investigated N AS
04/1988 Marines enter Panama P US
05/1988 Senate ratifies INF treaty P FP
03/1989 Senate rejects Tower nomination N O
05/1989 North convicted by Federal jury N AS
06/1989 First veto, of minimum wage bill -
08/1989 S and L Legis. signed by Bush -
10/1989 Dow Jones drops 190, 2nd lgst in history N NE
12/1989 Bush announces end of Cold War P FP
12/1989 US invades Panama P US
04/1990 Poindexter convicted N AS
08/1990 Iraq invades Kuwait, US and UN response P EN
10/1990 Bush vetoes Civ. Rights Act of 1990 -
11/1990 Red of conven weapons in Europe treaty P FP
01/1991 Desert Storm P US
03/1991 No fly zone in Iraq, shoot down Iraq plane P US
07/1991 Nuke reduction treaty (START) signed P FP
09/1991 US unilateral reduction in tactical nukes in Europe and Asia
P FP
10/1991 Thomas hearings, A. Hill revelations N
11/1991 Wofford beats Thornburgh for PA Senate N O
01/1992 Bush regurgitates in Japan -
04/1992 LA Riots N PR
| Variable | BH Event | Std Err | New Event | Std Err |
| Mean Model | ||||
| Constant | 5.68 | 1.11 | 5.19 | 1.15 |
| Inflationt-1 | -0.16 | 0.05 | -0.15 | 0.05 |
| Unemploymentt-1 | -0.29 | 0.91 | -0.51 | 0.77 |
| Watergatet-1 | -1.81 | 0.92 | -2.18 | 1.00 |
| Vietnamt-1 | -1.67 | 0.71 | -1.78 | 0.73 |
| Positive Eventt | 4.00 | 0.62 | ||
| Negative Eventt | -1.08 | 0.55 | ||
| Adverse Economic Eventt | -1.85 | 1.35 | ||
| Administration Scandalt | -0.10 | 0.45 | ||
| Domestic Accomplishmentt | 4.72 | 2.85 | ||
| Foreign Accomplishmentt | 1.93 | 1.14 | ||
| Escalationt | -1.08 | 1.79 | ||
| US-initiated Foreign Conflictt | 3.58 | 1.23 | ||
| Enemy-initated Foreign Conflictt | 4.15 | 1.02 | ||
| Approvalt-1 (l) | 0.91 | 0.02 | 0.91 | 0.02 |
| Moving Average (q) | -0.15 | 0.06 | -0.14 | 0.06 |
| s2 | 3.97 | 0.13 | 4.03 | 0.13 |
| Minimum | (SE) | Election | (SE) | Only Sig't | (SE) | |
| Mean Model Constant | 4.8662 | 1.0768 | 4.8685 | 1.0465 | 4.7231 | 0.7637 |
| Inflationt-1 | -0.1473 | 0.0501 | -0.1493 | 0.0512 | -0.1268 | 0.0490 |
| Unemploymentt-1 | -0.2658 | 0.9173 | -0.4878 | 0.8917 | -0.1415 | 0.8925 |
| Watergatet-1 | -1.5510 | 0.8541 | -1.5382 | 0.8850 | -1.5137 | 0.6562 |
| Vietnamt-1 | -1.4408 | 0.6639 | -1.5355 | 0.5625 | -1.4050 | 0.4898 |
| Positive Eventt | 3.6172 | 0.6513 | 3.3876 | 0.6466 | ||
| Negative Eventt | -1.0346 | 0.5390 | -0.9940 | 0.5515 | ||
| Adverse Economic Eventt | 3.3035 | -1.6945 | 1.7990 | |||
| Domestic Accomplishmentt | 3.4260 | 4.5318 | 1.1329 | |||
| US-initiated Foreign Conflictt | 3.4179 | 2.2394 | 0.9624 | |||
| Enemy-initated Foreign Conflictt | 4.4837 | 3.8259 | 0.8884 | |||
| Approvalt-1 (l) | 0.9197 | 0.0164 | 0.9204 | 0.4875 | 0.9215 | 0.0122 |
| Moving Average (q) | -0.1570 | 0.0629 | -0.1587 | 0.4327 | -0.1535 | 0.0615 |
| Variance Model | ||||||
| a0 | 3.5924 | 0.1478 | 3.5958 | 0.5715 | 3.5054 | 0.2265 |
| a1 | 0.0236 | 0.0076 | 0.0263 | 0.0076 | 0.0292 | 0.0083 |
| Re-election Yeart | 0.5894 | 0.4875 | 0.7423 | 0.4347 | ||
| Election Yeart | -1.2870 | 0.4327 | -1.5081 | 0.3596 | ||
| Percent Term Completedt | 0.4716 | 0.5715 | 1.0916 | 0.4905 | ||
| Administration Scandalt | -0.8902 | 0.2386 | ||||
| Domestic Accomplishmentt | -2.2618 | 0.8209 | ||||
2 The variance in presidential approval at any one time is closely related to the volatility of approval across time periods. Suppose that the researcher samples from a population where variance in approval is held constant. The volatility of approval - the poll-to-poll changes - will be proportional to the population variance, and a population with high variance in approval is likely to yield wilder fluctuations than one with lesser variance.
3 Notable exceptions areBurbach (1995), who categorizes foreign policy events into fiveseparate categories and who allows the size of a rally to vary withelite support and media coverage; and Marra, Ostrom, and Simon (1990),who categorized domestic and foreign policy events and presidentialtravel and speeches into a comprehensive list of ``determinants ofpresidential approval.'' (Marra, et al., pg. 596).
4 If events differ in the magnitudeof their effect, an overly coarse categorization will underestimatethe impact of some events and overestimate the impact of others.
5 Approval ratings after 1988 were collected from the Gallup Monthly. When approval ratings were collected less than monthly, the data were interpolated to monthly by linear averaging. When approval ratings were collected more than once a month, we chose the first observation collected in a month.
6 ARCH models have become popular in a variety of economic and finance applications, including capital and stock markets, and models of inflation, finance, and marketing (for a review, see Bollerslev 1992). Most applications involve dependent variables, such as exchange rates, where actors have had to make some forecast based on incomplete or uncertain information. Econometric models attempt to account for changing variance in these forecasts using the ARCH framework. If we conceptualize the approval rating as the ``current value'' the public places on the president's job performance, then the two applications are similar.
7 We did not have access to a complete set of the Galluppolls, as did Burbach (1995), and Ostrom and colleagues (Marra et al.,1990; Simon and Ostrom, 1989; Ostrom and Simon, 1989). By selectingonly the first poll in a month, we may underestimate the impact of anyparticular event on approval, but will not erroneously identify animpact where one does not exist (Ostrom and Simon, 1989, p. 369).
8 We drew upon Beck's (1989, 1991) model for themean. Our results are reasonably stable across model specifications,as displayed in the comparative table in the Appendix C: the significantvariables remain significant and of approximately the same magnitudeno matter what the specification of the model for the variance. Notealso that the model of the variance is largely insensitive to theselection of either the Brace and Hinckley event codes or to our moreelaborated set of codes.
9 Quarterlyeconomic data were provided to us by Neal Beck. These wereinterpolated to monthly figures using cubic spline curves. This methodof interpolation is virtually identical to linear interpolation whenthe time between observations is short and there are few gaps in thedata. The curvilinear nature of cubic splines are more appropriatefor the few instances where there are long gaps in the data.
10 The full list of events can be found in Appendix II of Brace and Hinckley (1991). When two positive or negative events occurred in the same month, we coded this as a single event (i.e. coded as ``1'' for that month). For the one month where a positive and negative event occurred (April 1952), we coded a ``0'' for both dummy variables. We coded additional events from 1988-1992, following the coding rules outlined by Brace and Hinckley. See Appendix A for more details.
11 In a series of papers, Ostrom and colleagues have developed an extensive categorization of presidential ``drama,'' from a twelve category listing of relevant outcomes, some of which were included in a final equation as a function of other, more distal variables; to others using the more conventional dummy variable approach (Ostrom and Simon, 1985); to the most recent eighteen category treatment of ``presidential drama'' (events, foreign travel, and speeches), depending on the primacy of the event and the domestic or foreign nature of the event (see Marra, Ostrom, and Simon 1990, pg. 599-606; Ostrom and Simon 1989; Ostrom and Simon 1986). While we developed our seven point scale independently, later reviews alerted us to the similarity of the approaches. In particular, our distinction between domestic and foreign policy events corresponds to the Marra, et al. distinction between president as foreign policy leader and as economic and domestic policy manager. Other distinctions that we make, such as U.S. vs. foreign initiated foreign conflict, are tailored more toward measuring changes in variance rather than the mean. Since the general categories are quite similar, and the results for the model of the mean replicate many others, we feel confident in continuing with our categorization.
12 We feel that it would bemisleading to give this scandal the same weight as other, morefrequent, but considerably more minor events (e.g. Sherman Adams's orDonald Regan's resignation). This will certainly depress the size of thescandal variable.
13 The percentage of independent identifiers was coded from assorted volumes of The Gallup Report. The fraction of party defectors and split ticket voters was coded from Stanley and Niemi (1995). These series were also interpolated to monthly figures.
14 The ARCH-EDL estimates in the model of the mean (the top panel ofTable ) are extremely close to Beck's original estimates(for a direct comparison, see Appendix B for the Beck modelestimated with these data).
15 There was only one event which qualified as a positive economic event, precluding a parallel test for the effect of good economic news.
16 Note that we are not suggesting that mean approval may not move up and down during an administration, only that the variance of approval increases over the duration of the administration, on average, in the post-war years.
17 Our results regarding foreign policy events dovetail with Burbach (1995). Burbach models variability among events and discovered (as we did) important differences in the size of rallies, depending on the kind of foreign adventure. The results just summarized, however, show that foreign policy events uniformly increase variance, regardless of whether they lead to large or small rallies or declines in approval ratings.