6. Voters are Mainly Sociotropic — Or Perhaps Not

Once the analysis of VP-functions moved into micro research, data increased dramatically to allow much stronger tests, but new interesting problems came up. The most intriguing was probably the sociotropic/egotropic controversy, where the two terms are defined in Table 7.14

Like the RP-pairs also ES-pairs exist in various data sets: The egotropic question is "how has your own economic situation developed in the last Z-period?" The corresponding sociotropic question is: "How has the economic situation of your country developed in the last Z-period?" The "economic situation" is sometimes replaced with "unemployment", and once more Z is typically either a quarter or a year.

The economically correct answer is surely that economic man is egotropic. Therefore, it was shocking when Kinder and Kiewiet (1979) demonstrated that the US voter was sociotropic. Several other studies — notably Lewis-Beck (1988) — have confirmed the result also for other countries, even after considerable refinement of the questions. So, for more than a decade there was no doubt that the main result in the literature was that voters are sociotropic, contrary to economic correctness.

Kinder and Kiewiet’s model was remarkably simple:


Table 7: Defining egotropic and sociotropic

The economic factor in the VP-function is:


what matters is the personal economy


what matters is the national economy

Here i is an index for the individual, Ei is the egotropic variable, Si is the sociotropic variable and Idi is a party identification variable.

The model of Kinder and Kiewiet was estimated on a cross-section. In an unusually aggressive article Kramer (1983) pointed out that this was not the right approach for the problem. In a cross-section the true sociotropic variable is constant, and hence unable to explain anything. What is estimated as sociotropic can only be different perceptions, Ni, of the same objective variable, Y:


Kramer did not see that the reformulation makes the economic correctness problem larger, not smaller! Surely, the rational voter perceives Y unbiased — that is, the perception error is white noise — and hence the coefficient 3 to Ni(Y) should be zero. Kinder and Kiewiet’s finding that the sociotropic term, 3, dominates the egotropic term, V, becomes even more mysterious. The most reasonable interpretation of the finding is thus that the different perceptions estimated must be due to different personal experiences, and hence that what is estimated as a sociotropic effect is really egotropic.

The next generation of models trying to come to grips with the egotropic/sociotropic distinction was introduced by Markus (1988). He reformulated (7b) into a mixed cross-section time series model, which includes a time index, t:


In Model (8a) the polled sociotropic variable is replaced with the "objective" one from the national statistical office. It can be compared with (8b) that should be almost the same as (7).

Model (8a) gives rather different results than (7). Now V and 3 become approximately the same on US data. Several studies have now estimated the various models for more countries. UK is covered by Price and Sanders (1994),15 Denmark by Nannestad and Paldam (1997a) and Sweden by Jordahl (2001). It appears that both V and 3 become significant, though not in a predictable mixture.

Thus the old agreement that voting is only sociotropic has given way to a more unclear picture, where voting is a mixture of egotropic and sociotropic factors. This is less economically incorrect than the old view, but a sizable sociotropic factor still appears in voting in most countries.

7. Voters are Uninformed about the Economy

Many polls have asked people about their knowledge of the economy. It appears that nobody has collected such polls systematically, and few studies have therefore been made comparing results.16 However, the results show rather decisively that voters know little about the economy.

In our Danish polls (see Nannestad and Paldam, 2000) most voters knew the number of unemployed within a few percentage points. They tended to know that the country had a balance of payments deficit and a budget deficit, both when the country had such deficits, and when it did not. When the two balances had the reverse sign, many mixed them up. Virtually nobody knew the sizes of the balances. Also, about f of the voters could not give an assessment of the inflation rate within 2 percentage points (that is, if inflation was 3% they were either outside the range of 1-5% or answered "don’t know").

However, there was one time during the four years covered, where knowledge increased substantially. This was around a general election. All of a sudden people knew that the balance of payments had changed from red to black figures.

The same type of result has been found about the EU: In European countries with no referenda institution, people know little about the European Union, but in countries with frequent referenda people know a lot more (see Paldam, 2001).

So, either people do seek information in connection with elections or they lean from watching the big show of an election/referendum campaign in TV. The fact that the information level goes up is worth to note as it explains why vote functions have a lower fit than popularity functions. It also explains why party popularities are normally much more volatile around elections than else.

Economic theory predicts that voters know what they need to know. The marginal benefits of information should be equal to the marginal costs:

MB(I) = MC(I) (the condition for rationality of  information) (9)

One way to define rational expectations is to demand that (9) is fulfilled.17 This is the definition used in the present paper.

The big problem surrounding (9) is what the benefits MB(I) are. Is it possible to argue convincingly that MB(I) is significantly larger than zero?

It is sometimes argued that people do need information on the macro level when they buy and sell shares and bonds (i.e., in connection with financing a house or a business) and when they buy and sell foreign exchange in connection with journeys abroad. But the theory of market efficiency — where many professional dealers are present — does effectively cut the link to information here. A (small) dealer in these markets can find no better information than the price.

The one occasion where people need information about the macro economy is when they vote at national elections and at their union, etc. Hence, we write:

MB(I) = MB(V)/MI (marginal benefit of economic information should be equal to the marginal improvement in the benefits derived from voting better) (10)

This brings us to the problems surrounding MB(V), the marginal benefit of voting. It is the problem known in the paradox of voting, where almost the same equation as (10) can be written:

MB(V) = MC(V) (the condition for rationality of voting) (11)

Much has been written about (11), but nobody has been able to argue convincingly that MB(V) differs significantly from zero. That makes all terms in (10) and (11) zero! However, we know that people do vote even if they do not know much about the economy. Surely, our friend economic man is in trouble. He should know nothing about the economy, and never vote. In short, the average voter behaves differently from economic man.

8. Two Ways to Rescue Economic Man

Two lines of arguments are often used to rationalize this apparent mess described in the last section. One is to argue that the cost of information, C(I), is small too. The second is to show that the cost of voting in an election, C(V), is large, and belongs to the class of social capital observations. Unfortunately, the two attempts are rather in the opposite direction.

C(I) is small too. This argument has been used by researchers such as Wittman (1995), Erikson et al. (2000) and Sanders (2000).18 They argue the voters know what they need to know, and that this is not much. What they need is to "grasp" what is going on, i.e., to have some feel for the way the economy is going. This feel is acquired from the media without really trying. If most experts look gloomy, then things are going badly, and if they look relaxed then things are going well. If a feel is enough then it is possible to argue that C(I) is almost zero as well. There is something in this argument, and Sanders does show that soft "feel" questions about the economy can be scaled to track the actual economy reasonably well.

If everything goes on at the extreme low-cost end of the scale, so that MB(I) = MC(I) = MC(V) = MB(V) = 0 which is almost zero, then things become a bit wooly. Also, it is well-known that people are unable to make utility assessments involving small numbers.

The other approach is to start from the large size of C(V). People do spend a lot of time looking at election campaigns in TV. Many go to some meetings, pay membership fees to political parties. Nearly everybody spends an hour driving to the election location, waiting in line and voting. In short, they have considerable costs participating in the national democratic process. Also, we know that parties find it worthwhile to spend millions on their campaigns. The total costs in time and money of a national election are likely to be something like 1% of GDP in the typical democracy.19 So we have to explain why so much is spent by the individual and the political system.

Voting is a social capital phenomenon It is a well-known observation that people in experiments play the cooperative solution much more than they should if they solve the game as good little economic men. The excess frequency with which the cooperative solution is played is a measure of the social capital or the mutual trust of the players. The key point to note is that trust does not need to be specific. It can be general. Many attempts have been made measuring general trust, and we know that it is much higher in some — generally more successful societies — than in others.20

In infinitely repeated games it is possible to uphold cooperative solutions, and it is arguable that society is a large number of games played with many players in different combinations over and over, seemingly without end. We do not bother solving all games in which we participate — and some of the everyday games are too complex to solve. So we develop rules of thumb standards, which can be termed trust. If a country succeeds in reaching a high level of trust then it is an advantage to everyone, as the cooperative solution becomes the one automatically reached in many situations.

The attempts to integrate these observations in standard theory are still going on. It seems likely that they may succeed, and then the paradox of voting may be solved at long last.

Given that this solution works, then we know that people do undertake considerable cost to follow the political scene and participate in the process. Given that these costs are so high, why not imagine that people also try to follow the economy? The fact that they are not so knowledgeable may simply be that it is difficult, as all those of us who teach the subject know.

It is interesting that the two attempts to save economic man are so different. In my judgement the second approach is the most promising, but it is not yet integrated into standard theory. However, both approaches agree that economic man is the wrong model for the job at hand.

9. Voters have a Grievance Asymmetry

Already the first popularity function study (Mueller, 1970) — discovered that voters react more to a negative economic event than to a corresponding positive one.

It was also found by Bloom and Price (1975) commenting on Kramer (1971). Then, for some time nobody looked for a grievance asymmetry, and the effect was forgotten.

Once the analysis of micro data became organized, it became possible to look for more effects. And, in the 1990s several studies looked carefully and found a rather strong asymmetry. See, e.g., Price and Sanders (1994) and Nannestad and Paldam (1997b). It appears that if e.g., unemployment increases by 1% (point), it has twice as large a negative effect as the positive effect of a fall in unemployment of 1% (point). The effect is thus large, and it is highly significant.

This has the consequence that if the economy moves from A to B in a straight line then the voter is more content than if it moves in a more roundabout way. In short, the variation around the growth path causes the government popularity to fall. Consider the average (Avr) and variance (Var) of a positive variable, i.e., a variable where MVP/MX > 0:


This is the standard formulation of risk aversion, so this is well integrated into economics, as long as it is prospective.

If the variables are retrospective, it is different. Relation (12) now changes from risk aversion to loss aversion. This is, in principle, an important change entering into one of the economic-man-problems discussed by Kahneman (1994) and other critiques of standard theory. The problematic aspect is that utility becomes path dependent. Costs are not sunk at all.

However, if expectations are stationary then it becomes unclear if we are dealing with risk aversion — which is perfectly nice theory — or loss aversion — which is a bad anomaly!

10. It Costs Governments Votes to Rule

The average government in a mature western democracy loses 2|% of the vote just by ruling (see Paldam, 1991; and Nannestad and Paldam, 2002). There is no difference between the average outcome and its standard deviation in such countries as the Netherlands, Sweden, Italy and Japan so it is a surprising fact. Also, it is heavily underresearched.21

Per definition the average government rules exactly as the rational voter should expect. So the economically correct prediction must surely be that the voter votes as before. It is hence a strange result that the average government loses a highly significant fraction of the votes. Three theories try to provide an explanation:

The oldest (from Mueller, 1970) is the coalition of minorities theory. It essentially says that the governments are formed when parties before elections manage to put together inconsistent coalitions. However, when they rule, the inconsistencies are reviled. So essentially the theory is that you can get away with unrealistic promises when in opposition. In average 2|% of the voters form unrealistic expectations at every election. This is a small irrationality, but it is a long-run fault, so it is not a "nice" theory, and it appears that no other evidence exists for the theory.

The second theory is the median gap model. It starts from a slightly amended version of the median voter theory. The pure form of the model suffers from the well-known problem that if both parties accept the policy of the median voter, they become perfectly alike and there is no reason to vote. To be distinguishable there has to be a gap around the median position between the voters. If there is a gap of size ( some of the voters will get as close as they can come to their ideal policy if the government changes at every election. This has all been worked out as a perfectly respectable model, and it is even possible to calibrate the model with reasonably looking parameters so as to explain the observed fact. Unfortunately, little corroborating evidence exists for the theory.

Finally, the third theory builds upon the grievance asymmetry just discussed. Imagine that the outcome is symmetrically distributed around the expected outcome. So, some variables improve and some deteriorate, but the gains and losses are equally large. If the reaction is asymmetric then there must be a loss of popularity in average. It is easy to show that the model produces the typical cost of ruling for reasonably sized VP-functions with the typical grievance asymmetry and realistically noisy economic outcomes.

Let us for a moment assume that the last theory is true. The average government loses votes because it rules and this causes governments to change. This destroys the evidence for the median voter theorem. In a two-party (two-block) system both parties (blocks) converge to 50% of the votes just because they rule when they are larger than 50%. The simplest explanation of the cost of ruling thus undercuts one of the key theorems of political economy.

11. Conclusion: Look at the Other End of the RE-Spectrum

Throughout the above survey it has appeared that the findings in the large VP-literature contradict the notions that the average voter behaves as does economic man of standard theory. That is, the symmetric, forward-looking agent, who takes everything relevant into consideration for his decisions. This is the fellow we constantly meet in the theoretical literature. However, the voters we meet "out there in the real world" do not optimize forward, but backward, and they have a short time horizon. Also, they have a strong grievance asymmetry, so that it cost votes to rule.

Section 7 argued that it would be irrational if the voter behaved as economic man! Voting is a decision where ignorance is rational, even if it is accepted that a "feel" for the economy is enough, we are still faced with complex and unsolved questions.

However, once you start from the notion that politics is a field with much uncertainty, and hence a short time horizon, things start to fall into some order. Under the circumstances it would be inconsistent, if the voters had long memories. This means the election cycles are out of the question, but it is consistent with the notion of partisan cycles. The shorter the time horizon the better does the past predict the future. Surely, within a few quarters nothing is likely to change very much.

The sizeable fraction of sociotropic voting — it is probably 25-50% — also makes a lot of sense when you ask what elections are all about. It does deal with the whole economy, not with the economy of Mrs Voter herself! So perhaps it is not so surprising that people give some consideration to the way they feel the government handles the economy when they vote.

Also, we do have a lot of results showing that the average citizen in all countries has considerable risk aversion (or, for that matter, loss aversion). The whole of the financial sector makes a perfectly good living out of turning risk pooling into a negative sum game for everybody else. So it is natural to expect a clear grievance asymmetry.

Once that is accepted, the cost of ruling follows. It is an important finding as it causes parties to change in power, and governments coalitions to converge to 50% of the vote. This happens irrespective of the median voter theorem and the minimum winning coalition theorem. Or rather it produces exactly the same observable facts in a much simpler way.

In short, it is worthwhile to take the findings in the VP-function literature seriously and use these findings for the development of a more realistic theory. Such a theory will also make it easier to explain the many suboptimal — or even crazy — outcomes we observe.

Next post:

Previous post: