Resumé

This article investigates the impact of macroeconomic conditions on aggregate political
support for governing parties in Sweden over the period 1967-1978. After reviewing survey
evidence on the relative salience of economic and fiscal issues to the Swedish electorate, a
novel, dynamic model of political support is presented. The model incorporates the ideas
that voters evaluate economic performance relatively rather than absolutely, and that
governments' mass political support is basedon their cumulative performance records. The
empirical results supply estimates of the impact of unemployment, inflation, the real
income growth rate and the income effects of direct tax and transfer policies on political
support. The evidence suggests that the responses of political support to reasonable movements
in macroeconomic conditions are large enough to give economic management a
pivotal role in electoral shifts.

1. Introduction

Although economic issues probably were less important in Swedish political life in the 1960s than during the immediate postwar period (or during the 19305), public concern about employment, inflation and taxation had by no means expired. Table 1, reporting survey data on policy priorities of the Swedish electorate in 1960, 1973, and 1976, illustrates the persistent salience of these macroeconomic issues. During the 19705, questions of social reform commanded the electorate's foremost attention, which is not surprising in view of the steady expansion of social programs and public expenditures. Sixty-one percent and forty-seven percent of those responding to the policy priorities questions in the 1973 and 1976 surveys, respectively, identified explicitly economic policy and taxation as salient political issues, and because social reform issues have important implications

---

* This research was supported by National Science Foundation Grant SOC 78-27022 (to Douglas A. Hibbs, Jr.) and by a Danish Social Science Research Council grant (to Henrik Jess Madsen). We are grateful to Nicholas Vasilatos for computational assistance and to Christine Aquilino for typing the manuscript.

for state spending and, hence, touch directly upon general economic concerns,
these numbers may well understate the extent to which economic factors
influenced voters' evaluation of the government.

Indeed, more detailed survey evidence supports this conjecture. Table 2 presents the electorate's view of the importance of employment, inflation, taxes, and energy for their vote in the 1976 election. The data show rather convincingly that these policy areas were of crucial importance. Fully 82 percent of the respondents saw the taxation issue as either one of the 'most important' influences or as a 'rather important' influence on their vote, whereas about 52 percent attached equivalent importance to the energy question. We do not wish to argue against the decisive role played by energifrågan for the outcome of the 1976 election (a point that has been emphasized by Swedish election specialists);l nevertheless, in view of

the data in Table 2, as well as other evidence,2 we want to caution readers
against minimizing the electoral impact of economic concerns.

The 1976 election aside, have macroeconomic conditions contributed in a systematic way to the electorate's support for Swedish governments? Since 1967 SIFO has regularly probed the vote intention of a representative sample of the Swedish electorate,³ and we have used this time-series (monthly surveys aggregated to quarterly observations) to analyze the influence of macroeconomic performance on fluctuations in mass political support for Swedish governments. The paper is organized as follows. Section 2 presents a statistical model of qualitative political choices which formalizes the idea that the discrete support/nonsupport responses elicited by surveys are reflections of underlying, continuously valued voter sentiments about the incumbent political party or bloc ranging from strongly positive to strongly negative.

Section 3 of the paper develops several specifications for the way in which voters evaluate governments' economic performance and reports the empirical results. Four measures of economic performance appear in the political support equations: the rate of inflation; the rate of unemployment; the growth rate of per capita real disposable income; and the difference, or gap, between the growth rate of post-tax, post-transfer per capita disposable income and pre-tax, pre-transfer per capita earnings. The latter variable is designated tax gap'. It essentially measures the wedge between the growth streams of final disposable income and original market income opened up by state tax and transfer policies.

The specification for voters' evaluations favored by the evidence incorporates the idea that people evaluate economic performance relatively rather than absolutely by comparing current economic conditions to those 'expected'from experiences in the recent past. This specification also embodies the notion that a government's current support is based on its cumulative relative performance record;

however, since the present relevance of information conveyed by past performance
decays over time, it is assumed that voters discount backward in time the
retrospective relative performance streams.

The regression coefficient estimates for this model indicate that deviations of the unemployment rate from its 'expected' level exert the strongest influence on variations over time in support for the governing party or bloc. Inflationary bursts also have important political consequences, whereas the estimated impact of tax gap is small and that of the real disposable income growth rate is negligible.

The concluding section of the paper reviews the results and develops their larger
implications.

2. A Model for Qualitative Political Choices

Although public opinion polls typically force people to make discrete, qualitative responses - in our case whether the respondent intends to vote for (or prefers) the incumbent party or bloc - in principle a voter's preference is not a discrete Tor' or 'against' phenomenon, but instead is a matter of degree falling on an underlying continuum ranging from strong positive to strong negative feelings about the governing party or bloc. Therefore the dependent variable in the model is a continuously valued, unobserved index of support for the governing party/bloc at time t, Yf. Y* is determined by the equation

(1)

where f(Z) represents an evaluation function for the vector of economic performance
variables, Z, which is described ahead; and ut is an independently distributed
random disturbance.

Y* is unobserved; the SIFO survey data reveal only voters' discrete party preferences or vote intentions. Hence we need a model that maps the discrete choices in the SIFO surveys onto the unobserved, continuously valued support index. Let the observed survey responses be designated by the binary variable Y_jt:

(2)

Since this paper focusses on movements through time in aggregate political support, we shall assume that individuals react homogenously to government performance, f(Z), and, therefore, that the Y_it reflects the underlying continuously valued support index such that

(3)

where c is a critical threshold.

It follows that the probability (P) of survey responses supporting the government
at time t are

(4)

and (1 - P) gives the probability of support for nonincumbent parties (or of abstention). The probability of supporting the incumbent party therefore hinges on the value of c - f(Z) and the distribution of the random variable u. The point is illustrated by Figure 1.

The above implies that P_t may be regarded as a cumulative distribution function. Any appropriate distribution for u will yield a well behaved probability function. It is convenient, however, to assume u logistic (which differs trivially from the normal distribution) with mean zero and unit scale parameter, which implies the probability function

where we have modified the conventional Berkson (1955) logistic function model along the lines proposed by Amemiya and Nold (1975) to include the random variable v_t to take account of omitted independent variables. v_t is assumed to have constant variance 02.o². As illustrated by Figure 2, equation (5) means that the response probabilities monotonically approach one as f(Z) + v_t goes to +°° and monotonically approaches zero as f(Z) + v_t goes to -°°.

Replacing the true probabilities (PJ with the aggregate proportions
N
observed in the survey data (P't = I Ylt/Nt) and manipulating (5) gives the
i=l
logit estimating equation

(6)

where e_t = ln(P'_t/l - P't) - ln(P_t/l - P(), and without loss of generality cis arbitrarily set equal to zero.⁴ It can be shown that (6) implies a regression equation with heteroscedastic residuals (the variance of the t^th residual is a² + (N_tP_t(l - Pi))"¹) and, therefore, we employed the two-step, weighted least-square estimator developed by Amemiya and Nold (1975).

The next section describes alternative specifications of the performance
evaluation function, f(Z), and reports the empirical results.

3. Performance Equations and Empirical Results

The regression analyses are based on quarterly observations over the period 1967:1 - 1978:3, which includes part of the Erlander government, and all periods of the Palme and Fälldin governments. For the reasons reviewed in the last section, the dependent variable in the regression equations is ln(P'_t/l - P't) (the logit of P't), where P'_t is the proportion of the SIFO sample indicating a preference or willingness to vote for the governing party(ies) in quarter t.⁵ As indicated in the introduction, the economic performance variables are the unemployment rate, the inflation rate, the per capita real disposable income growth rate, and the gap, or difference, between the per capita growth rates of final disposable income and original market income, which is induced by the tax-transfer system (tax gap).6 Each regression also includes a separate intercept constant for Social Democratic and Bourgeois governments. These constants are essentially ignorance terms that pick up factors unrelated to economic performance advantaging one or the other party or bloc.

Three specifications of the performance evaluation function, f(Z), were entertained. The first is a rather naive model which assumes that voters' political support for governing parties responds only to contemporaneous macroeconomic performance:

(7)

where aha2 denote party/bloc intercept constants; Zisa vector representing the
economic performance variables with associated coefficients b; and heteroscadiscity
weights have been omitted for convenience.

Equation (7) is essentially the same as the model applied by Jonung and Wadensjö (1979) to monthly data on untransformed support proportions and, not surprisingly, it yields results not too dissimilar to the findings of their study.7 The coefficient estimates for this model reported in Table 3 show that the unemployment, inflation and tax gap variables are correctly signed, but only the unemployment estimate is statistically significant. Hence, there is little doubt from these results that political support for governing parties is quite sensitive to movements in unemployment. Notice that the real income growth rate enters the regression with a perverse (negative) sign; however, the magnitude of its coefficient is negligible and statistically insignificant.

Column 2 of Table 3 gives estimates for a more plausible model in which current political support is based on an exponentially weighted moving average of economic outcomes during the entire course of a government. A government's contemporaneous standing with the electorate therefore depends on its cumulative economic record, rather than solely on economic conditions in the current period. The estimates for this model are derived from the equation

(8)

where: q is an index of each government (1 = Erlander, 2 = Palme, and
3 = Fälldin);

g is the rate of decay of the distributed lag weights,
02sg<l;Aq:=:+l during the qth government and zero
otherwise;
qt is the number of periods that the qth govemment has been in
power;
and, as before, Z represents the economic performance variables
with associated coefficients b.

As Equation (8) implies, the regression coefficient b gives the total impact of a movement in Zon the political support index ln(P'_t/l - P't). The total impact of a sustained change in Z is distributed over the life of a government according to the geometric lag sequence

The estimate of the nonlinear coefficient g was obtained manually by searching
over the parameter space and choosing the value minimizing the error sum of
squares.

The estimates for equation (8) in Table 3 show that it explains variations in the underlying political support proportions slightly better than equation (7); the correlation of the actual proportions and the fitted proportions implied by the logit model is 0.63 as opposed to 0.615 for the previous equation. However, the substantive implications are basically the same: only unemployment has a sizeable and statistically significant impact on political support. The persistent political impact of unemployment fluctuations revealed by the results for equations (7) and (8) causes little surprise. Although unemployment compensation and other transfer

payments cushion many of the unemployed against significant income losses,⁹ only about half the unemployed are members of an insurance fund. For at least a small fraction of the electorate, then, unemployment means real hardship.lo However, regardless of financial compensation, unemployment imposes psychological costs, that although difficult to measure quantitatively, are, nonetheless, likely to be sharply felt. Also, in addition to households touched directly by some form of unemployment or underemployment, an even larger number will be aware of unemployment among relatives, friends, neighbors and workmates. Moreover, since unemployment represents foregone real output and underutilized human resources, it confers costs on society as a whole as well as on the individuals affected directly.

The discomfort associated with inflation, on the other hand, arises primarily from unanticipated movements in prices rather than from the inflation rate per se. Anticipated inflation presumably is reflected in wage claims and settlements and in other nominally priced contracts that are not explicitly indexed and, therefore, does not generate great dissatisfaction in the electorate. Voters probably also give governing parties little credit for rates of growth in real disposable income, or for growth in disposable income relative to market income, that simply reinforce customary experience or prior trends. Therefore, the small, statistically insignificant coefficients for the inflation and income variables reported in Table 3 for equations (7) and (8) have a sensible interpretation.

The third performance evaluation function that we have investigated takes explicit account of actual economic performance in relation to customary or expected performance. This model formalizes the idea that voters react to sharp deviations of actual economic outcomes from their customary or expected levels, rather than to economic outcomes per se. If, for example, the contemporaneous inflation or unemployment rates exceed what people are accustomed to, political support will decline. Conversely, support is enhanced when unemployment declines briskly to historically low levels or inflation decelerates sharply. The same holds for the difference between the customary and actual tax gap or real income growth rate, but with opposite effects on political support. This reasoning implies the equation

(9)

where

and other terms are as defined earlier.

As in equation (8), the relative performance deviations in equation (9), Z - Z*, are cumulated over the life of each government with past deviations weighted g_fc. Customary or expected performance outcomes, Z*, which serve as the benchmarks for voters' reactions to actual outcomes, Z, are generated autoregressively. In other words, the Zf are formed from linear combinations of previous outcomes with coefficients Cj obtained by regressing Z_t on ZZ_{t _i,} Z,_₂,..., Z_t j.^{111 1} Notice that customary performance, Z*, is not fixed: the equation for Z* in (9) means that performance standards are relative rather than absolute and adjust dynamically over time. Hence, any sustained rate of inflation, unemployment, real income growth or tax gap will eventually become 'customary' to the electorate, satisfying (Z_t - Zf) =0. Political rewards and penalties are conferred only by fluctuations in economic conditions producing non-zero deviations of Z from Z*, that is, deviations of current conditions from a linear combination of conditions in the recent past.

The parameter estimates for equation (9) are shown in Table 3, column 3. The correlation of the predicted and actual SIFO proportions shown at the bottom of the Table (r = 0.7) indicates that this equation does a better job of explaining the underlying empirical data on political support than equations (7) and (8). All parameters have the anticipated sign and, with the exception of the real income growth rate coefficient, are statistically significant. It is clear from these results, then, that sharp upward movements in the rates of unemployment and inflation (positive deviations of Z from Z*) yield losses of support for the governing party or political bloc, and that unexpected upward movements in disposable income growth rates relative to market income growth rates (tax gap) produce increases in the government's support in the electorate. It should be noted that large gaps between disposable and market income growth rates generally imply expansions (if the gap is positive) or contractions (if the gap is negative) of cash transfers relative to collective consumption. Nonzero values of this term may also stem from reductions (if Z - Z* is positive) or increases (if Z - Z* is negative) of direct rates of personal taxation that are not accompanied by changes in transfers. The significant positive coefficient for tax gap variable in equation (9) means, therefore, that voters reward governments for unexpected increases in transfers relative to collective consumption expenditures or, alternatively, react favorably to sharp reductions in direct taxation that do not adversely affect the flow of transfer spending.l2 Finally, the empirical results indicate that in the presence of the unemployment, inflation and tax gap variables, the real disposable income growth rate alone does not have a sizeable impact on the government's standing in the SIFO polls.

The estimate of the lag weight decay parameter g in equation (9), which
defines the distribution over time of the ultimate impact of a sustained

value of (Z - Z*), is about 0.8. This implies that the electorate's current
vote intentions are iniluenced by relative performance outcomes extending many
periods back through time. As the normal electoral period is
11
three years (12 quarters), we have at the election quarter (l-.8)/812) I
k = 0
g(Z- Z*)tt_k and, therefore, the percentage of the election quarter impact
of a persistent unit deviation of Z from Z* that is felt by the kth lag is given
by (l-.8k+l)/(l-812).13 Hence, at the typical election period contemporaneous
relative economic performance (k = 0) accounts for about 21 percent of the economy's
impact on voting preferences, relative performance over the previous year (k
= 0, 1, 2, 3) picks up about 63 percent of the economy's election quarter impact,
and relative performance over the preceding two years (k = 0,1,..., 7) represents
about 89 percent of the economy's effect on the governing parties' electoral support.

4. Implications for Observed Political Support

The empirical results discussed above indicated that equation (9) is the most satisfactory specification of the impact of economic performance on electoral support. However, the regression coefficient estimates pertain to the logits, ln(P'_t/l - P't), whereas practical interest centers on the consequences of relative economic performance for the percentages (or proportions) of the electorate supporting governing parties. Since the survey proportions, P'_t, are a nonlinear function of the corresponding logits, ln(P'_t/l - P't), the precise effects of practical interest are not immediately obvious from direct inspection of the parameter estimates. Therefore, to illustrate the practical political consequences of fluctuations in the significant economic variables, we have computed the changes in the percentage of the electorate supporting the governing parties following unfavorable movements of Z relative to Z* that are sustained for one and four periods prior to the end of the Erlander, Palme and Fälldin regimes. To facilitate comparisons of the estimated effects of movements in the economic variables, changes of one standard deviation in the (Z - Z*) terms are considered. Table 4 displays the results.¹⁴

Adverse fluctuations in unemployment relative to expectations that are based on recent past experiences, clearly exhibit stronger influence on electoral preferences during the Erlander, Palme and Fälldin governments than unfavorable shifts in the inflation or tax gap variables. Averaged over all governments, a standard deviation increase of unemployment beyond the expected level and lasting

only one period produces a decline of about 1.3 percentage points in the governing parties' mass political support. Sustained a full year (4 quarters), the same adverse increase (acceleration) of unemployment relative to expectations on average generates losses of about 3 percentage points in governments' electoral support.

The political consequences of unfavorable, relative changes in the inflation and tax gap variables are smaller, with the former variable being about twice as important as the latter. Sustained just one period, adverse movements in inflation and tax gap relative to voters' recent past experiences decrease political support on average by 0.9 and 0.4 percentage points, respectively. The political penalties are larger if the same unfavorable relative changes in these variables are sustained over four quarters: the loss associated with inflation averages 2.7 percentage points and with tax gap it averages 1.31 percentage points.ls

In view of the extremely competitive electoral politics of the postwar
Swedish party system, the losses of political support attributed to reasonable
movements in the macroeconomy in Table 4 are hardly trivial. Although

the 'Socialist bloc' (Social Democrats and Communists) commanded a comfortable lead during the 19605, subsequent electoral margins were much narrower. As Table 5 shows, averaged over all elections beginning in 1952, the margin of victory separating the Socialist and Bourgeois blocs has only been 1.6 percent of the vote.¹⁶ Even though partisan preferences are obviously also based on more enduring characteristics of political life than those incorporated explicitly in our model, our results indicate that macroeconomic management can play a pivotal role in electoral shifts.

The comparatively strong electoral effects arising from movements in the unemployment rate illustrated in Tables 3 and 4 obviously reflect the salience of the employment issue in postwar Swedish politics. While the anticyclical policies pursued by the Social Democrats in the 1930s provided only marginal relief from the severe recession, they skillfully exploited its symbolic significance and became identified with full employment and prosperity. The specter of a repetition of the mass unemployment of the 1930s under a Bourgeois government became the underlying theme in the Social Democrats' electoral strategy, and it proved to be an advantageous mobilizing device. However, it gradually receded in importance as postwar cohorts entered the electorate and as a result of Bourgeois parties' successful efforts to convince the electorate that they also were committed to full employment.l7 As a result, by the late sixties the relative tightness of labor markets was the preeminent measure of governments' economic performance, which is what our estimates suggest. A subsequent paper will pursue the implications of these results for the formation of macroeconomic policies in Sweden.

---

NOTES

1. See, for example, Petersson (1978) and Zetterberg (1977).

2. For example, in the post election sample of the 1976 election survey, 33 percent of the respondents who reported changing party between the 1973 and the 1976 election rated taxation as one of the absolutely most important issues for their vote in 1976. Thirty-one percent identified the nuclear power problem as the most important; twenty percent thought of inflation in that way, while fifteen percent emphasized the employment issue.

3. Until 1969 the question asked was Which party do you prefer today?' Since 1970 the phrasing has more explicitly focussed on vote intention: Which party do you consider voting for at the next election?' A similar survey was initiated by Statistiska Centralbyrån in 1972. Since the SIFO surveys extend further back in time and are available more frequently, we have used the SIFO data time-series.

4. Since c is necessarily embedded within the intercept constants of f(Z) evaluation equation (see ahead), nothing is lost by setting it to zero here. The proof that eq. (5) implies eq. (6) is as follows. exD fZ) For simplicity let P' = -—-, which is the form of eq. (5). Then, 1-F-l- fPg) hence 1 + exp (Z) P exp (Z) P' = exp (Z), from which it follows that ln(P'/l - P') = Z, which is the form of eq. (6).

5. During the Erlander and Palme governments P'_t is the proportion supporting the Social Democrats; during the Fälldin government it is the sum of the proportions supporting each of the bourgeois parties in the government.

6. Hence tax gap_t = (Zi_t-Z2t) where Z_lt is the growth rate of per capita personal disposable income, and Z2t is the growth rate of per capita market earnings. An appendix giving data sources and data series is available from the authors upon request. All rates of change are formed ln(Z_t/Z_t-i) • 400, i.e., quarter-on-quarter differences of the logs expressed at annual rates.

7. Jonung and Wadensjö (1979) also estimate models with a lagged endogenous variable to pick up (via the Koyck transformation) distributed lag effects. However, the Durbin's h statistics they report are significant, indicating that the disturbances are autocorrelated and, therefore, that the regression coefficient estimates are biased and inconsistent.

8. Since the sum of the geometric lag weight sequence Qt ~ 1 l-g^qt I y^k is ,_ , premultiplying the lag distribution by (l-g)/(l-g^qt) simply acts as a k = 0 [ g normalizing constant that insures that the lag weights sum to 1.0 at every period during the life of a government. This is why b gives the total impact of a sustained movement in Z and the lag weights define its distribution backward through time.

---

9. For example, the annual disposable income of a male metal worker unemployed for one month in 1978 amounted to97%ofhis full employment annual income. If unemployed for six months, he would have 90% of his full employment annual disposable income (cf. J. Persson-Tanimura 1980, table 1).

10. The Swedish Low Income Commission identified unemployment as accounting for about 50% of the poverty gap.

11. Since the Cj coefficients in the autoregressive equations generating Z* are likely to evolve through time, these equations were reestimated for each government using observations up to the beginning of that government. The optimal autoregressive equations for each variable were: Autoregressive Equations Generating Expected Economic Performance (Z§ (standard errors in parentheses) standard constant ZZ_{t -j} ZZ_{t -2} ZZ_{t -3} error of the regression Erlander (1951:4-1955:4) Inflation 2.535 0.246 0.068 3.24 (0.72) (0.13) (0.13) Unemployment 0.299 0.667 0.154 0.25 (0.09) (0.12) (0.12) Real Disposable Income 3.754 -0.371 -0.259 5.96 (0.95) (0.14) (0.16) Taxgap 1.403 -0.755 -0.361 -0.18 7.64 (1.06) (0.14) (0.17) (0.18) Palme (1951:4-1969:3) Inflation 2.535 0.191 0.099 3.21 (0.66) (0.12) (0.12) Unemployment 0.294 0.638 0.194 0.25 (0.09) (0.12) (0.11) Real Disposable Income 3.531 -0.324 -0.229 5.95 (0.83) (0.12) (0.12) Taxgap 1.72 -0.683 -0.317 -0.21 7.22 (0.94) (0.13) (0.15) (0.15) Falldin (1951:4-1976:3) Inflation 1.947 0.204 0.157 0.234 3.56 (0.733) (0.10) (0.10) (0.103) Unemployment 0.256 0.745 0.117 0.24 (0.08) (0.10) (0.09) standard constant ZZ_{t -\} ZZ_{t -2} ZZ_{t _3} error of the regression Real Disposable Income 5.201 -0.516 -.461 -0.483 9.05 (1.17) (0.11) (0.11) (0.11) Tax gap 1.027 -0.513 -0.17 -0.199 7.16 (0.75) (0.11) (0.12) (0.12)

---

11. Since the Cj coefficients in the autoregressive equations generating Z* are likely to evolve through time, these equations were reestimated for each government using observations up to the beginning of that government. The optimal autoregressive equations for each variable were: Autoregressive Equations Generating Expected Economic Performance (Z§ (standard errors in parentheses) standard constant ZZ_{t -j} ZZ_{t -2} ZZ_{t -3} error of the regression Erlander (1951:4-1955:4) Inflation 2.535 0.246 0.068 3.24 (0.72) (0.13) (0.13) Unemployment 0.299 0.667 0.154 0.25 (0.09) (0.12) (0.12) Real Disposable Income 3.754 -0.371 -0.259 5.96 (0.95) (0.14) (0.16) Taxgap 1.403 -0.755 -0.361 -0.18 7.64 (1.06) (0.14) (0.17) (0.18) Palme (1951:4-1969:3) Inflation 2.535 0.191 0.099 3.21 (0.66) (0.12) (0.12) Unemployment 0.294 0.638 0.194 0.25 (0.09) (0.12) (0.11) Real Disposable Income 3.531 -0.324 -0.229 5.95 (0.83) (0.12) (0.12) Taxgap 1.72 -0.683 -0.317 -0.21 7.22 (0.94) (0.13) (0.15) (0.15) Falldin (1951:4-1976:3) Inflation 1.947 0.204 0.157 0.234 3.56 (0.733) (0.10) (0.10) (0.103) Unemployment 0.256 0.745 0.117 0.24 (0.08) (0.10) (0.09) standard constant ZZ_{t -\} ZZ_{t -2} ZZ_{t _3} error of the regression Real Disposable Income 5.201 -0.516 -.461 -0.483 9.05 (1.17) (0.11) (0.11) (0.11) Tax gap 1.027 -0.513 -0.17 -0.199 7.16 (0.75) (0.11) (0.12) (0.12)

12. See Hibbs and Madsen, 1981, for some comparative evidence on this point.

11 13. The sum of the finite geometric series (l-0.8)/8¹²) I•gk k = 0 I—o 8 rnok+l iq Q^+l is -_— ' • -_—-h_— =-— . Since by construction the lag weights sum to 1.0 at every 1~0.o¹² l~u.o 1~0.8¹² period, this gives immediately the percentage of the ultimate impact felt by the kth lag.

14. The losses are inferred by comparing the support percentages predicted by equation (9) from the actual, historical economic data to predictions generated by the equation holding the (Z - Z*) variables one standard deviation above or below their historical values. The end of the Erlander and the Falldin governments of course did not follow elections. Nevertheless, for illustrative purposes the impact calculations in Table 4 might be interpreted as the implied losses of support on 'election days'. The Erlander government was considered to have been in office since 1964:3 and the standard deviations were therefore calculated for a period starting at this date and ending in 1978:3. As the economic variables are expressed as deviations from autoregressively formed expectations, a four quarter sustained change implies accelerating rates of inflation, unemployment and tax gap. The mean and standard deviations of the three (Z - Z*) performance variables are mean standard deviation inflation 1.85 3.81 unemployment 0.05 0.23 tax gap -0.05 6.35

15. The difference in the simulated losses of political support between the three administrations partly reflect differences in the lengths of time in office. As the Falldin government ruled for only eight quarters (compared to Erlander's 72 quarter tenure and Palme's 28 quarter incumbency), the economic effects tend to be larger, due to the g lag weight scheme, for the Bourgeois coalition. The estimated electoral losses basedon equation 8 (for appropriately calculated standard deviations of the original variables) are similar to those presented in Table 4. The implied electoral losses averaged over all three governments are (in percentage points): sustained for 1 quarter 4 quarters inflation -0.53 -0.86 unemployment -1.36 -2.22 tax gap -0.57 -1.09 Notice that movements in the rate of unemployment and in tax gap variable are relatively more important here.

---

15. The difference in the simulated losses of political support between the three administrations partly reflect differences in the lengths of time in office. As the Falldin government ruled for only eight quarters (compared to Erlander's 72 quarter tenure and Palme's 28 quarter incumbency), the economic effects tend to be larger, due to the g lag weight scheme, for the Bourgeois coalition. The estimated electoral losses basedon equation 8 (for appropriately calculated standard deviations of the original variables) are similar to those presented in Table 4. The implied electoral losses averaged over all three governments are (in percentage points): sustained for 1 quarter 4 quarters inflation -0.53 -0.86 unemployment -1.36 -2.22 tax gap -0.57 -1.09 Notice that movements in the rate of unemployment and in tax gap variable are relatively more important here.

16. When evaluating Table 5 in conjunction with Table 4, it should be remembered that the dependent variable in the regressions for'Socialist'_governments was basedon the survey proportions supporting the Social Democrats only.

17. See, for example, Särlvik 1977.

REFERENCES

Amemiya, T. and Nold, F. 1975. 'A Modified Logit Model,' Review of Economics and
Statistics 57, 255-257.

Hibbs, D. A., Jr. and Madsen, H. J. 1981. 'Public Reactions to the Growth of Taxation
and Government Expenditure,' World Politics, in print.

Jonung, L. and Wadensjö, E. 1979. The Effects of Unemployment, Inflation and Real
Income Growth on Government Popularity in Sweden,' The Scandinavian Journal of
Economics 81, 343-353.

Persson-Tanimura, I. 1980. 'Vad kostar arbetslösheten?', Ekonomisk debatt 8, 107-115.
Petersson, O. 1977. Väljarna och valet 1976, Stockholm: Liber-Förlag/Allmänna Förlaget.

Petersson, O. 1978a. The 1976 Election: New Trends in the Swedish Electorate,' Scandinavian
Political Studies 1 (nos. 2 and 3), 109-121.

Petersson, O. 1978b. Valundersökning 1976: Teknisk Rapport, Stockholm: Liber-Förlag/
Allmänna Förlaget.

Särlvik, B. 1977. 'Recent Electoral Trends in Sweden,' in K. H. Cerny (ed.), Scandinavia
at the Polls, Washington, D.C.: The American Enterprise Institute.

Zetterberg, H. 1977. The Swedish Election of 1976,' paper presented to the World
Association of Public Opinion Research, Uppsala, August 14-19.