Consider the challenge of a political party that loses an election, or a sequence of elections, in a two-party system. Party activists must wonder: what should the party change to win the next election? Standard spatial theories (Downs Reference Downs1957) say the party’s platform must have been too far from the median voter’s preferred policy, and that to win, the party should moderate its platform, bringing it in line with the median voter’s wishes.
However, sometimes a losing party does the opposite: instead of moderating, it doubles down, moving away from moderate policies and embracing radical positions. For instance, in the United Kingdom, after Labour lost the 2010 and 2015 elections under mainstream candidates (Gordon Brown and Ed Miliband), it chose a far-left new leader (Jeremy Corbyn). Labour soon went on to suffer heavy losses of seats in Scottish, Welsh, and local council elections in 2016 and 2017 and its electoral prospects became so poor that the UK’s Tory government called an early election, anticipating a near certain win. Footnote 1 Labour similarly followed its 1979 General Election defeat by lurching left toward its greatest electoral defeat in modern times in 1983. In the USA, following a defeat in the 1960 election, the GOP chose one of its most extremely conservative senators (Barry Goldwater) as its candidate, leading to a landslide loss in 1964.
We ask why rational, office-motivated parties would choose extremism and near-certain defeat when moderation offers better expectations of victory. Non-instrumental theories would account these choices as expressive (Brennan and Hamlin Reference Brennan and Hamlin1998). Extreme policy platforms are more satisfying to purist partisan factions (Roemer Reference Roemer2009, ch. 8), even if they spell electoral doom. We suggest a different, purely instrumental, answer: weak parties go extreme to increase the probability of winning subsequent elections, even at the cost of losing the immediate one.
Consider a two-party system in which parties enjoy policy-specific reputation (or valence) advantages, as discussed in Krasa and Polborn (Reference Krasa and Polborn2010, Reference Krasa and Polborn2012, and Reference Krasa and Polborn2014). In particular, assume that one party “owns” the mainstream ideas (Petrocik Reference Petrocik1996), in the sense that the voter trusts its handling of the mainstream policy better. The other party can cede this mainstream ground, investing instead on an a clearly distinguishable alternative policy, developing quality proposals (Hirsch and Shotts Reference Hirsch and Shotts2015) around this alternative. We call this alternative “extreme” merely because it falls outside the mainstream. By persevering on these position-specific investments, the party acquires a policy-specific valence advantage on this alternative policy position. We call this choice tactical extremism.
Tactical extremism is a last-resort strategy. As long as voters continue to prefer the mainstream policy, it leads to electoral defeats. However, negative economic outcomes induce voters to update negatively on the virtues of the mainstream policy. If voters become disillusioned, and wish for an alternative, tactical extremism pays off: the party that had chosen tactical extremism enjoys the valence advantage in providing such an alternative. Two assumptions are key: party valence is policy specific and endogenous; and voters’ future policy preferences are uncertain.
A disadvantaged party chooses tactical extremism under two conditions: confidence in the currently mainstream policy is weak, so that voters interpret a negative outcome as a policy failure and not as bad luck; and a party’s credibility or competence on a particular policy is important to voters.
We show that a global downturn increases the chances of tactical extremism only if confidence in the mainstream policy is initially low. Countries with a sufficiently strong confidence on mainstream policies will not experience more tactical extremism after a downturn. While tactical extremism may seem an unambiguously negative phenomenon, we show that in some circumstances, it increases welfare, and voters would like more tactical extremism than we observe in equilibrium.
Related literature analyzes the rise of outsider candidates (Buisseret and van Weelden Reference Buisseret and van Weelden2017; Karakas and Mitra Reference Karakas and Mitra2017) and populist parties (Guiso et al. Reference Guiso, Herrera, Morelli and Sonno2017). Mainstream policy-motivated parties may also adopt non-median policies (Calvert Reference Calvert1985), which can be welfare enhancing (Bernhardt, Duggan, and Squintani Reference Bernhardt, Duggan and Squintani2009). Such polarization increases in the proportionality of the electoral rule (Matakos, Troumpounis, and Xefteris Reference Matakos, Troumpounis and Xefteris2016). Office-motivated parties sometimes choose a left-of-center policy to signal that they are not conservative (Acemoglu, Egorov, and Sonin Reference Acemoglu, Egorov and Sonin2013) or to change voters’ ideologies (Prato Reference Prato2018). And incumbents can benefit from announcing extreme positions because voters discount their announcements (Bawn and Somer-Topcu Reference Bawn and Somer-Topcu2012).
Closest to our work, Aragonès and Palfrey (Reference Aragonès and Palfrey2004) show that a weak candidate who could not win at the center can win with some probability by leaning moderately away from the center to differentiate itself from its stronger rival; in contrast, we explain why a party that can win at the center chooses to make losing more likely by lurching to an extreme platform.
An extended literature review, discussion of motivating historical examples, rigorous formalization of the model and results below, extensions and robustness checks with further results, and all proofs, are available in the online Appendix to this letter.
THE MODEL
Setup. Consider a two-period model of electoral competition with two purely office-motivated parties and one strategic representative voter. In each period t ∈ {1, 2}, parties A and D compete in an election (an extension to an infinite horizon is in the Appendix). Parties seek to maximize the sum of the probabilities of being elected over the two periods. We consider a policy space X = {e, m}, where m represents an orthodox standard policy and e represents an unorthodox experimental policy. We refer to m as the mainstream policy and to e as the extreme policy.
In each period t, before the election, each party
j ∈ {A, D}
simultaneously announces a platform
$x_t^j \in X$
, which is the policy that the party will implement in
period t if it wins office. The voter observes
$\left( {x_t^A,x_t^D} \right)$
and votes for either A or
D. The winning party W
t
∈ {A, D} implements its
announced policy
$x_t^{{W_t}}$
.
Let o
t
denote the economic outcome in period t.
Economic outcomes are uncertain. A time invariant state of Nature θ ∈ Θ
= {e, m}, determines which policy is
more likely to deliver a good outcome. In each period
t, the probability of a good outcome is πh ∈
(0, 1) if the implemented policy
$x_t^{{W_t}}$
matches the state θ, and πl ∈ (0,
πh) otherwise. We refer to the policy that coincides with
state θ as the correct policy and to the other policy
as the wrong one. We refer to policy m
as the mainstream one because all agents ex-ante agree
that m is the policy most likely to be correct.
We model policy-specific valence (or
competence) by assuming that whether the government
implements its chosen policy competently affects the utility of the
voter. We denote party j’s competence in period 1 as a
function of
$x_1^j$
by
$c_1^j\left( {x_1^j} \right)$
, and party j’s competence in period 2
as a function of
$x_2^j$
given
$x_1^j$
by
$c_2^j\left( {x_2^j|x_1^j} \right)$
. Competence in period 2 is also a function of the
previous platform because acquiring competence on a given policy
requires time to build the necessary expertise. We assume that in period
1,
$c_1^A\left( m \right) = c$
and
$c_1^A\left( e \right) = 0$
, where
$c \in \left( {0,{1 \over 4}} \right)$
is an exogenous parameter, observed by all players. On
the other hand, party D has no competence
$\left( {c_1^D = 0} \right)$
on either policy in period 1. Thus, party
A has an exogenous competence advantage on the
mainstream policy. The intuition is that both A and
D are traditionally mainstream parties with an
asymmetry: one of them is perceived as more competent than the other at
delivering mainstream policies. We highlight this deliberate asymmetry
by henceforth referring to party A as the
advantaged party and party D as
the disadvantaged party. In the second period, for
party A, we assume that
$c_2^A\left( {m|m} \right) = c_2^A\left( {e|e} \right) = c$
and
$c_2^A\left( {e|m} \right) = c_2^A\left( {m|e} \right) = 0$
and for party D that
$c_2^D\left( {e|e} \right) = c$
and
$c_2^D\left( {e|m} \right) = c_2^D\left( {m|e} \right) = c_2^D\left( {m|m} \right) = 0$
. The intuition is that party A owns
the mainstream policy position, in the sense that it enjoys a
policy-specific valence advantage that D cannot match
in two periods but which is nevertheless lost if A ever
abandons this policy position. On the other hand, the extreme policy
position is up for grabs, and a party that commits to it for both
periods gains a competence advantage on it.
Footnote 2
We also model non-policy valence (or
charisma) by assuming that ε
t
represents the voter’s idiosyncratic preference for party
A in period t. The random
popularity shock ε
t
captures non-policy attributes that affect the voter’s
preferences. In each period t, ε
t
is drawn independently from a uniform distribution over
$\left[ { - {1 \over 4},{1 \over 4}} \right]$
.
Timing. The state of Nature θ ∈ {e,
m} remains unknown throughout the game, but a
common prior
$\mu \equiv \Pr \left[ {\theta = m} \right] \in \left( {{1 \over 2},1} \right)$
is common knowledge. The probabilities (πl,
πh) are also common knowledge. The realization of the
non-policy shock ε
t
is unknown at the beginning of period t. At the
start of period 1, party A has a competency advantage
on policy m. This party always prefers to announce
m, so for simplicity we assume that
A is restricted to announce
$x_1^A = m$
. Party D chooses
$x_1^D \in \left\{ {e,m} \right\}$
. After these party choices, all players observe
$\left( {x_1^A,x_1^D,{\varepsilon _1}} \right)$
. Then, the voter chooses a vote in
$\left\{ {A,D,\emptyset } \right\}$
. For each party j, if the voter votes
j, then j wins, while if the voter
abstains
$\emptyset$
, the winning party is randomly chosen with equal
probability. The winning party implements its policy. Then the economic
outcome o 1 is realized and observed by all
players (see the period timeline in Figure 1).
FIGURE 1. Period t timeline, for t∈{1,2}.
At the start of period 2, after observing the economic outcome
o 1, all players formulate a posterior
belief μ* about which policy is correct. This revision from the prior to
the posterior belief may justify a change in the advocated policies, so
both parties are able to formulate a new platform for period 2. Each
party j simultaneously chooses
$x_2^j \in \left\{ {e,m} \right\}$
. Then
$\left( {x_2^A,x_2^D,{\varepsilon _2}} \right)$
are observed, the voter chooses the winning party,
which implements its policy, and the economic outcome
o 2 occurs.
Utilities. Parties are purely office motivated. They maximize the expected number of periods in office, without time discounting.
The voter optimizes period by period, myopically. Footnote 3 In each period t, and for each party j, the voter calculates the expected utility that it would attain if she elects party j. This expected utility is computed as the sum of three terms: the expected economic performance under party j (given the voter’s beliefs), the policy-specific valence of party j, and the non-policy valence of party j. The voter then optimizes for the period by voting for the party with the highest expected utility.
Solution concept. We assume that parties are strategic and sequentially rational, while the voter votes in each period for the party that delivers her a higher expected period utility, conditional on her beliefs. Beliefs follow Bayes’ rule and are consistent. We provide a formal definition of belief consistency and of the equilibrium concept in the Appendix.
We will say that there is Tactical Extremism (TE) if party D chooses the extremist platform in the first period. We assume throughout that
$$\mu\,\, \ge\,\, {1 \over 2} + {{1 - 4c} \over {8\left( {{\pi _{\rm{h}}} - {\pi _{\rm{l}}}} \right)}} \,\,\equiv\,\, \bar{\mu },$$
which guarantees that in the first period, the voter prefers a party with a mainstream policy so that choosing the extremist policy will lead to certain defeat in the first election. This condition makes it harder for TE to obtain.
RESULTS
We next describe equilibrium behavior.
At the voting stage in the first period, voters prefer the mainstream policy, and they only vote for a party with a mainstream platform. So the implemented policy in the first period is the mainstream one.
Consider the policy choice in the second period. The observed economic outcome in the first period provides a signal about the mainstream policy. If it is good, then the second period posterior belief that the mainstream policy is correct is high, and both parties choose the mainstream platform.
In contrast, a bad economic outcome in the first period sends a negative signal about the mainstream policy and, depending on the (ex-ante) level μ of confidence in this mainstream policy, it alters the platform choice in the second period. With little initial confidence in the mainstream policy, the posterior belief after a bad signal becomes low enough that both parties react by choosing an extremist platform in the next period. For intermediate levels of confidence, party A sticks to the mainstream platform while party D proposes the extremist one. For high enough confidence, both parties choose the mainstream platform.
Consider the incentives for party D to engage in Tactical Extremism (TE), i.e., to propose the extremist policy in the first period. Choosing platform e, party D foregoes any chance of winning the first election, so it only has any incentive to propose e in the first period if doing so helps to win the second period election. Choosing policy e in the first period only helps if party D chooses e again in the second period, and then it enjoys a policy-specific valence advantage c on policy e. In the second period party D chooses policy e if and only if the prior confidence μ in the mainstream policy is not too high. Therefore, holding other parameters constant, TE only yields any advantage if the prior confidence in the mainstream policy is not too high. The magnitude of the advantage is equal to the competence parameter c. This leads us to our main result.
Remark 1 Tactical Extremism. The disadvantaged party engages in Tactical Extremism (TE) whenever the voter’s initial level μ of confidence in the mainstream policy is sufficiently low and competence is sufficiently important (c is high). Footnote 4
Competence has two complementary effects. Suppose competence matters more (c increases). In the first period, for party D the probability of winning with the mainstream platform is reduced—its disadvantage is bigger—and this reduces the cost of investing in extremism. Second, in the second period, conditional on a disappointing outcome from the mainstream policy, the investment in expertise on the extremist policy is more valuable and this reinforces the incentives for TE in the first period.
Comparative statics on the effect of the importance of competence (c) or confidence in the mainstream policy (µ) are straightforward: if the importance of competence increases, TE arises for a greater range of prior confidence (µ) in the mainstream policy; if the confidence μ increases, then TE arises for a smaller range (of very high values) of competence c. Footnote 5
We next consider comparative statics on the effect of changes in the economic environment, described by the parameters (πh, πl). Note that πh − πl is the probability that the outcome is good if and only if the government implements the correct policy, and 1 − πh is the probability of bad luck, in the sense that the outcome is bad regardless of the policy choice.
Assume that the underlying environment deteriorates to
$\left( {\rpi _{\rm{h}}^\prime ,\rpi _{\rm{l}}^\prime } \right) = \left( {{\rpi _{\rm{h}}} - \Delta ,\;{\rpi _{\rm{l}}} - \Delta } \right)$
for some Δ > 0. A first effect is that a
negative economic outcome becomes more likely, favoring TE. A second,
indirect, effect is that if a bad economic outcome occurs, the
probability that the policy choice is to blame is lower (because the
probability of bad luck has increased to 1 − πh + Δ). When
bad outcomes are more common, a bad outcome sends a weaker negative
signal about the policy choice, and after a bad outcome, the posterior
confidence in the mainstream policy shrinks less. This second effect
deters TE.
If confidence in the mainstream policy (μ) is initially low, the strength of the signal doesn’t matter: even a weak negative signal sinks the voter’s confidence in the mainstream policy, so the first effect dominates. Whereas, for a fixed high μ, only a strong negative signal induces the voter to prefer the extreme policy, so the second effect dominates.
Remark 2 Impact of Exogenous Factors on Tactical Extremism. If the prior confidence (μ) in the mainstream policy is low, a deterioration of the underlying economy increases the incentives for TE. Whereas, if the prior confidence (μ) in the mainstream policy is high, a deterioration of the underlying economy decreases the incentives for TE.
If μ is low, with a worsened economic environment, the minimum value of competence c necessary for TE to occur decreases, so the parameter range for which TE occurs expands. Whereas, if μ is high, the minimum value of c for TE to occur increases, so the parameter range for which TE occurs shrinks.
WELFARE
Naive intuition might suggest that Tactical Extremism is detrimental to the voter. However, there are trade-offs.
In the first period, the effect is unambiguously negative because TE reduces choice. In this period, the voter wants a party with a mainstream platform. Without TE, the voter has two such parties to choose from to select the one with highest aggregate valence. With TE, party D essentially takes itself out of the running for this period, so the voter is worse off.
In the second period, conditional on a bad realization of the first period economic outcome and conditional on the prior confidence μ on the mainstream policy not being too high, TE is beneficial to the voter. Under these two conditions, the voter has lost confidence in the mainstream policy and wants policy e. With TE, there is one party (party D) offering such platform with high competence; without TE, no party would be competent on this extreme policy.
We say that TE is welfare enhancing if the voter is better-off under TE than under a default in which both parties propose in each period the policy that is in expectation best in that period. The parameter region for which TE is welfare enhancing does not coincide with the region for which TE is an equilibrium phenomenon.
Remark 3 Welfare implications. If the initial confidence in the mainstream policy (μ) is low, voters would prefer more TE than obtains in equilibrium (too little TE). On the other hand, if the initial confidence in the mainstream policy (μ) is high, voters would prefer less TE than obtains in equilibrium (too much TE).
If there is sufficient uncertainty about the mainstream policy’s effectiveness, the voter would like to “insure” against the risk that the extremist policy is better, by assuring that one party has competence on it. Party D is more reluctant to go extreme because she fully internalizes that TE brings a loss of the probability of winning the first election, whereas the voter does not care about the identity of the winning party. With strong confidence in the mainstream policy, the voter does not worry so much about such risk and focuses on the fact that TE forces her to vote for party A in the first period. In sum, TE benefits the voter if she has enough doubts about the mainstream policy that she values having a good alternative ready, in case it might be wanted in the second period.
DISCUSSION
We have identified conditions under which one of the two parties in a two-party system adopts an extreme policy for tactical reasons. We predict that a party is more likely to engage in this tactical extremism when
-
(a) its reputation for policy-specific competence in delivering standard, mainstream policies is poor and
-
(b) voters’ confidence in mainstream policy prescriptions is not too high, and this confidence is at least somewhat likely to dissolve in the future.
With this prediction in mind, we revisit our motivating example: the Labour party lurch toward the extreme left in 2015. The financial crisis of 2007–09 and its handling by the Gordon Brown’s Labour government had battered the reputation for managerial competence of Labour’s pro-market centrist wing. At the same time, the crisis and subsequent austerity drives had undermined voters’ confidence in mainstream expert policy prescriptions. Conditions for Labour to tactically choose extremism were more favorable after 2010 than at any other time since the early 1980s, when Labour had last tried extremism, under Michael Foot. In the 1980s, extremism did not bring success to Labour. We wonder: will it work now?
Our theory would say: “maybe, but more likely it will not.” Tactical extremism is a risky strategy that appeals only to a party at a substantial disadvantage. By going extreme, the party condemns itself to an immediate electoral defeat in hopes of a future electoral gain that may not materialize. The party bets against the policy prescriptions that are most likely to be correct, betting that subsequent events will prove these prescriptions wrong in the eye of voters. Since these prescriptions are more likely to be right than wrong, tactical extremism is more likely to fail than to succeed.
SUPPLEMENTARY MATERIAL
To view supplementary material for this article, please visit https://doi.org/10.1017/S0003055418000758.
Comments
No Comments have been published for this article.