Previous Work

Because most of the states that adopted the initiative in this period were agrarian, western, and mostly white, historians have traditionally answered this question by claiming that the initiative was most likely to be adopted in “younger states with a strong history of populism and antimonopoly sentiments, more homogeneous [white] populations…, more radical (but weaker) political parties, and stronger interest groups” (pg 334). Smith and Fridkin contend that these conclusions have been based on deeply flawed logic and methods.¹

In particular, scholars have erred in asking why some states adopted the initiative and others did not; such an approach focuses on voter demand for direct democracy. But since the initiative is usually adopted via legislative referendum, in which legislators invite voters to choose whether to adopt the initiative, scholars should instead be asking why legislators in some states chose to offer the initiative to voters, while legislators elsewhere did not.²

Theory and Findings

Smith and Fridkin argue that legislators were most likely to offer the initiative when states experienced close partisan competition. When a minority party gains tenuous control of the legislature, or when a majority feels that it might become the minority soon, then legislators gain an incentive to weaken the legislature by empowering the median voter. This is their main causal argument, which is supported by an event history analysis.

Other conditions also play a role. In particular, legislatures in newer states are more like to offer the initiative than legislatures in older states. The authors attribute this difference to the weaker party organizations in younger states–organizations that might have less ability to rein in independent-minded legislators. Likewise, legislators were more likely to offer the initiative if there were more third-party legislators or if there was a strong interest group pressure in a particular state for instituting the initiative.

Contrary to previous work, the authors do not find any evidence of a contagion/diffusion effect. Nor do they find that initiative adoption was more common in racially and ethnically homogeneous states. Significantly, they find that dummies for the western and southern region are not statistically significant in their models; controlling for other variables apparently eliminates the regional bias in initiative adoption.

Critique

This paper’s methodological rigor makes it a welcome addition. As the authors point out, most previous work on this topic has been somewhat impressionistic and ad hoc. Event history analysis is the right way to study adoption of the initiative. Their shift from studying adoption to studying legislative support for the initiative is very smart.

That being said, a few things seem off to me. First, why did initiative adoption largely cease after 1918? Imperfect as previous work may be, its emphasis on western culture and populism seems to have an answer to this question. Smith and Fridkin don’t so much as speculate; they simply restrict their analysis to 1898-1918. But if some omitted variable caused initiative adoption to cease in 1918, is it possible that the same omitted variable, if measured, would change the results in Smith and Fridkin’s analysis?

Also, it seems that the role of governors may be downplayed. The authors start well (p 339) by listing several Democratic and Republican governors who pressured their legislators to pass an initiative reform; they even note that Republican Hiram Johnson of California made this a major goal. Yet inexplicably, the authors “control” for these considerations by including a dummy for whether the governor belonged to a third party. They provide no justification of this measure which, unsurprisingly, is not statistically significant–leading them to conclude (without basis) that gubernatorial action plays little role in adoption of the initiative. But if we believe that executives can pressure legislators by “going public” or through other means, then we should be looking more at the role governors may have played.

Third, they take no account (empirically) of variations in initiative institutions. Initiatives vary widely in (1) how hard they are to use and (2) how easily the legislature can ignore them (Bowler and Donovan 2004). Why did some states (California, Oregon) pass powerful initiatives, but other states (Utah, Idaho) passed weak ones? The authors acknowledge this variation, but do not seek to explain it. My speculation: The “party competition” variable that the authors discuss might be two different things. If a minority (perhaps allied with third parties) seizes control for a short while, or if a majority genuinely fears becoming a minority, then we might see an effort to create a truly powerful initiative. But if a majority sees its power slipping and wants to do something popular to build up support, then we might see a weak initiative created purely for its symbolic value. Thoughts?

Every study has weaknesses, and the flaws in this one are by no means fatal, nor are they necessarily more numerous or severe than what is typical for an article in a top journal. But I think they point to some avenues for further research.

It is one of the happy incidents of the federal system that a single courageous state may, if its citizens choose, serve as a laboratory; and try novel social and economic experiments without risk to the rest of the country.

These claims have inspired a literature on “diffusion,” mostly focused on the American states, that has asked whether there is evidence that the states do, in fact, learn from one another. For the most part, the existing literature has concluded that diffusion occurs regularly among the American states, although there is some disagreement as to the mechanism.¹

By contrast, Volden, Ting, and Carpenter argue that none of the existing work (including their own previous work on diffusion, apparently) has provided any evidence of policy diffusion. The methods and assumptions used in previous research cannot differentiate between innovation (isolated experimentation by myopic states) and diffusion (learning from experiments in other states).

Overview of the Model

To correct this problem, the authors present two formal models of experimentation. Both begin from the same basic setup: States have policy makers (legislators, bureaucrats, whatever) that can be placed along a unidimensional ideological line (i.e. we can classify the state as conservative or liberal). Within a particular issue area, there is a status quo policy and a proposed experimental policy. Each policy has two characteristics. First, it can be placed along the ideological line, and this “spatial” characteristic is common knowledge–that is, everybody agrees as to which policy’s goals are more liberal or conservative. Second, each policy has a “valence”–that is, each policy might be more or less effective at reaching its stated goals.

Although each proposal’s “spatial” characteristics are assumed to be common knowledge, “valence” is known only for the status quo–the experimental proposal’s valence is unknown. Thus, policy makers have a choice: They can stick with the status quo (with known valence) or they can switch to the experimental policy (with unknown valence). If they choose to experiment, then in “period two” (e.g. the next legislative session), when the experimental policy’s valence is known, they can choose to stay with the new policy or revert to the old one.

After setting up this basic model, the authors derive two models from it. The first is a decision-theoretic model that assumes each state exists in isolation; states may innovate similar policies, but there cannot (by definition) be diffusion. The second is a game-theoretic model that assumes each state can learn from policy experiments in other states; either innovation or diffusion can occur.

The decision-theoretic model

In this model, states do not have the option of learning from other states’ experiments.

In an extremely liberal state, the policy makers will choose the most liberal policy proposal, regardless of expectations about valence. (Likewise for extremely conservative states). But in moderate states, policy makers will balance valence against ideology. A moderate conservative would prefer an efficient but liberal policy over an inefficient but conservative policy; a moderate liberal would prefer an efficient but conservative policy over an inefficient but liberal policy. These are the conditions under which innovation occurs.

Thus, we would expect to see only moderate states experimenting. If they learn that the policy is inefficient, then they would revert to the previous policy–the one that is ideological preferable.

The game-theoretic model

In this model, states do have the option of learning from other states’ experiments.

If states can learn about a proposed policy’s valence by observing policy experiments in other states, then the incentive to experiment drops. Experimentation is risky; if you can learn from others’ mistakes rather than having to make those mistakes yourself, then why experiment at all? As such, those policy makers willing to experiment will fit into a narrower ideological range than those willing to experiment in the decision-theoretic model.

Implications for the Literature

Previous work on “diffusion” has not appreciated the differences between these two models. All of the evidence for policy diffusion presented in previous work can be explained in terms of the decision-theoretic model. In order to conclude that diffusion actually occurs, we must find evidence of behaviors that are predicted by the game-theoretic model and NOT by the decision-theoretic model.

The literature has presented five different causal mechanisms to explain “diffusion,” but all five mechanisms can be explained with the decision-theoretic model:

• Walker says some states are inherently more disposed to innovate than others.
• Gray says diffusion happens when states face similar policy problems.
• Others say that diffusion happens when neighboring states are ideologically similar.
• Still others say that diffusion happens when any states (neighboring or not) are ideologically similar.
• A final argument is that diffusion happens when policy advocates take their arguments to multiple states.

As the authors put it: “Much of the empirical work to date has not adequately distinguished [game-theoretic] learning-based policy diffusion from [decision-theoretic] myopic individual adoptions.”

The authors conclude by listing specific empirical implications of their model that future research should evaluate in order to determine whether or not diffusion actually occurs.

Critique

Much of the work on diffusion was published in the 1960s and 1970s. I suspect that modern political scientists have hesitated to take up the question again because they thought it was settled. This article demonstrates persuasively that the question is far from settled. This is the article’s most important contribution–to point out that we do not yet have any evidence that states do (or do not) act as “laboratories of democracy” that learn from one another. This question is normatively important; if states do not learn from one another, then we lose an argument for devolution.

However, I was less than satisfied with the authors’ empirical suggestions. They conclude by pointing out several ways that we can use their theory to empirically determine whether states follow the decision- or game-theoretic model. Unfortunately, these empirical implications are extremely nuanced and may be difficult (or nearly impossible) to apply in practice.