Digital Information and Policy Innovation in US StatesAbstract

We examine the role of broadband subscriptions and policy innovation in US states. Diffusion scholars have long recognized the role of information in the spread of policy ideas. We argue that digital information access has made it easier for states to observe other actors, increasing the amount of policy information and making states more innovative. At the same time, we argue that growth of broadband internet has disrupted previous existing flows of information by decreasing the importance of geographic proximity and creating more nationalized information networks. We estimate a pooled event history analysis comparing 2000-2016 to the last two decades of the 20th Century and find that states with higher broadband subscriptions are more innovative overall and less reliant on contiguity for policy prescriptions. The growth of digital information has altered the diffusion network allowing states to become more innovative and less reliant on geography when evaluating policy ideas.

Introduction

The rapid spread of broadband internet has fundamentally transformed American society over

the last two decades. The internet has made it easier to communicate across vast distances,

leading many to say the world is getting “smaller.” People, organizations, groups, and lawmakers

can more easily learn about what others are doing across the country and the world. The changes

brought about from the spread of the internet are comparable to the invention of the printing

press, telegram, and telephone (Chadwick 2017; Silver 2012). Communities with higher levels of

internet connectivity are more economically dynamic, civically engaged, and interconnected than

areas that have not yet integrated themselves in the knowledge economy (Mossberger, Tolbert

and Franko 2012; Moretti 2012). Tracking the diffusion of over a hundred policies over the 20th

century, Boushey concludes that “mass media and communications technology has acted as an

important trigger for positive feedback cycles in policy diffusion” (2010, 173). Over the last two

decades the adoption of broadband internet has altered nearly every aspect of American society,

and may impact the spread of policy ideas across the states.

Technology expands the distance that information can travel, shortens timelines,

broadens networks for diffusion of information and mobilization. We argue states with greater

connectivity should have more information-rich environments. Broadband represents an

infrastructure for information (as the old terms says, an information superhighway). It is a

resource for innovation. Yet the literature has not measured or analyzed the role of digital

information--broadband internet--in policy diffusion.

Research in policy diffusion and innovation has implicitly or explicitly acknowledged the

central role of information in the spread of policies (Walker 1969; Savage 1985; Boushey 2010;

Mossberger 2000). States must be aware of what policies other states are adopting in order to

1

imitate, learn, or compete (Shipan and Volden 2008), and citizen exposure to other state policy

decisions is a key influence on state’s decision to innovate (Pacheco 2013). Additionally, the

states rely upon a robust interest group network to identify policy prescriptions for emerging

problems (Garrett and Jansa 2015). The influential work on punctuated equilibrium is an

information-based theory of policy change that conceptualizes political systems as “information-

processing instruments” (Baumgartner and Jones 2005, vii). Information flows to citizens and

elites have an influence on the diffusion of policies. The internet has dramatically increased the

amount of information policymakers receive, and we argue that states with higher broadband

adoption should be more innovative. Changes in information and communications technologies

should speed up diffusion processes.

A half century after Walker’s landmark study, digital media is the primary platform for

information access in the 21st Century. Walker (1969) documented cases of policy diffusion, or

the spread of policy ideas, among the American states as far back as 1870. Since that time, the

growth of the mass media and improved communications through internet have opened new

possibilities for the traffic in policy ideas. Today the diffusion of information can reach new

audiences at lightning speeds. Rising from less than 5% of the population with a home

broadband connection in 2000 to 85% in 2018 (combining wired or satellite home broadband

65% and cell phone only 20%, Pew 2018), the past two decades have witnessed a dramatic

transformation of information access via digital media.1 See Figure 1 showing the percent of the

population with a home broadband connection over time. Examining trends and disparities in

broadband access (subscriptions) is critical for understanding the past and the present of local

communities in the United States, and their needs and resources to address both innovation and

1 See figure 1 for data on broadband access, See figure 2 for the distribution of broadband across the states at different time points

2

inequality in the rapidly unfolding knowledge economy. Broadband, or high-speed internet, “has

become vital to almost every segment of the Nation’s social and economic fabric,” according to

the National Broadband Research Agenda (NTIA/NSF 2017, 3), affecting state governments as

well as individuals. Economic activity, political and civic participation and governance all

depend on networked flows of information and communication online. We expect states with

richer digital information access will be more likely to adopt new policies across the spectrum of

issues over time, as broadband makes the flow of information much easier and faster.

We use the new State Policy Innovation and Diffusion Database (Boehmke et al 2019) of

thousands of state policy adoptions and their diffusion from 1980-2016 and find digital

information access (i.e. percent of the population with home broadband subscription) increases

the probability of states innovating. We also find that broadband connectivity has altered the

flows of information states use when making policy decisions. Broadband’s ability to connect

distant communities has reduced the importance of geographic proximity in the spread of

information. Legislators in Mississippi can easily learn about what policies Alaska has adopted,

and interest groups can rapidly share policy information nationwide at a very low cost. We find

that while contiguity still positively predicts policy adoption, broadband reduces its importance

in policy innovation. Broadband internet adoption has altered how information flows between

the states and should be incorporated when understanding why some policies spread and others

do not.

Policy Diffusion and Innovation

Political scientists have long studied what causes some policies to diffuse widely and

others to not (Walker 1969; Gray 1973). Berry and Berry’s (1990) introduction of event history

analysis to diffusion scholars opened a new wave of interest as scholars could use internal and

3

external characteristics to model diffusion. This methodological innovation led to a large growth

in single policy studies that provided important insights to the predictors of policy diffusion

(Grossback, Nicholson-Crotty and Peterson 2004), although scholars have noted that

increasingly single policy studies are leading to diminishing returns (Boehmke and Skinner

2012). Recently, there has been a large growth in the number of studies that take a large-N

approach to measuring policy diffusion, often using dozens or hundreds of policies to

systematically identify trends patterns of policy diffusion (Boushey 2010; Boehmke and Skinner

2012; Kreitzer 2015; Karch, Nicholson-Crotty, and Bowman 2016; Caughey and Warshaw 2016;

Boehmke et al 2018). This trend has enabled researchers to identify consistent diffusion patterns

across policies and study systematically what drives diffusion.

As diffusion research has developed, scholars identified four primary mechanisms of

diffusion--imitation, learning, competition, and coercion (Shipan and Volden 2008). While the

first three mechanism represent the horizontal spread of policies across peer actors (between

nations, states/provinces, or cities) coercion represents a vertical process, such as the federal

government requiring states to adopt a policy. We focus on the three horizontal mechanisms to

understand how peer actors interact with respect to policy innovation. Imitation occurs when

states look to states with similar attributes for policy ideas, regardless of policy outcome.

Conservative states may look to conservative states for policy ideas from co-partisans, or states

may look to other states with similar population sizes or levels of wealth for policy ideas because

they policymakers assume similar states will face similar problems and policy solutions.

Learning, however, occurs when a state looks to the policy’s success or failure when

making decisions to adopt a policy. State similarity/dissimilarity is not as important when

learning occurs, because learning focuses on the policy’s effects, not the actor adopting the

4

policy. Success can be defined in a number of ways. The policy may achieve a substantive gain,

such as reducing poverty, increasing education outcomes, or variety of other socio-economic

benefits. Alternatively, co-partisans may look to see if a policy leads to partisan success.

Regardless of their effect on the budget, do tax cuts lead to incumbents winning reelection at

greater rates? Are spending cuts politically feasible? Politicians can learn from other states if a

policy is successful and are more likely to adopt that policy if it is deemed a success. The last

mechanism is competition. Competition occurs when states want to gain a competitive edge over

others. They may lower tax rates or adopt new economic development policies outperform

neighboring states, and the desire for gaining a competitive edge can spur policy innovations.

Despite the growth in research studying diffusion mechanisms, there has been little agreement on

to how best operationalize each mechanism (Gilardi 2016).

Despite difficulty distinguishing between mechanisms, scholars have identified factors

that increase the probability of policy innovation. Perhaps the most persistent finding in the

diffusion literature is that policies are more likely to diffuse across neighboring states (Walker

1969, Gray 1973; Berry and Baybeck 2005). States can more easily observe what policies

neighboring states are adopting, and citizens near state borders are likely exposed to neighboring

state policy changes (Pacheco 2013). At the same time, the advent of new methodological tools

and larger data sets of policy adoption have allowed scholars to take more systematic approach

to understanding diffusion and innovation. Researchers have identified latent policy diffusion

ties between the states (Desmarais et al 2015) that look very different from the contiguous

diffusion networks previously identified. These latent networks are composed of previous

contacts on policies adopted in the past. Other researchers have emphasized the role of outside

interests in the diffusion process. Garrett and Jansa (2015) use a network analysis of bill text to

5

identify how states not only emulate each other when adopting policies, but also that they borrow

substantially from interest groups when crafting legislation.

The diffusion of innovations literature has long measured access to information in

alternative ways, from Rogers (1962, 2003) who argued that individuals who were early adopters

were more cosmopolitan to the policy diffusion literature. Walker (1969) identified urbanization

as a factor in early adoption, and this is in itself an expression of the presence of information

networks. Glaeser (2011) has argued that large cities are more innovative because of the dense

networks of information they support, with knowledge spillovers across firms and sectors. An

educated workforce has long been identified as a factor in economic innovation (Moretti 2012)

and in the context of policy diffusion, legislative professionalization is commonly a predictor of

innovation. Educated populations, legislative professionalization and urbanization can all be

viewed as expressions of a networked information environment. Broadband subscriptions in the

population is a more direct and observable measure of information environments.

Despite the variety in methodological and theoretical approaches to understanding

diffusion, the central component is typically the role of information exchanges in policy diffusion

(Boushey 2010; Nicholson-Crotty and Carley 2018). Whether competing, learning, or imitating,

states use information about what other states are doing when they innovate. Information can

play a number of roles. Previous adopters send information to other states about the ideological

location of a policy (Grossback, Nicholson-Crotty & Peterson 2004). Partisans can learn that a

policy is politically viable and are more likely to adopt a policy previously adopted by

copartisans (Butler et al 2017). Information about experiences elsewhere reduce perceptions of

risk and uncertainty in adopting a new policy (Walker 1969; Rose 1991). Information flows are

not constrained to communication between the states but can also emerge from interest groups

6

and policy advocates pushing for states to adopt new policies (Mintrom and Vergari 1998, Balla

2001; Garett and Jansa 2015). Information-based theories of policy change include Baumgartner

and Jones (2005) punctuated equilibrium. Boushey’s (2010) research tracking the diffusion of

133 policies over three eras from 1900 onward demonstrates that over time there have been more

marked punctuations of rapid diffusion. Later policies are more likely to spread non-

incrementally (Boushey 2010, 60-61). Boushey suggests that changes in information and

communications technologies should speed up diffusion processes, but he does not directly

measure broadband use. Some policies that diffuse rapidly are politically salient and more reliant

on public opinion (i.e. morality policies or populist measures such as term limits) (Boushey

2010, 159). Although these policies pre-dated broadband, they indicate that public awareness of

information can matter for the adoption of new policies. Additional research has found that

contiguity’s role in diffusion has decreased in the last few decades (Mallinson 2019), with a

potential explanation being social media and digital information facilitating contagion effects

across large geographic areas. Information plays a vital role in the spread of innovations and

states’ ability to find policy prescriptions for emerging problems.

Our analysis conceptualizes states as networked information environments, with multiple

linkages both inside and outside their borders, and the ability to look near and far for ideas and

experience. States often have multiple sources within national information networks, where the

experience of other states is synthesized in research from think tanks, professional journals,

reports from federal agencies, and proposals and arguments from interest groups (Mossberger

2000; Mossberger and Hale 2002). Broadband adoption represents a measure of the information

environment in a state, including its ability to access information from outside its boundaries.

While policies vary in their diffusion patterns, those that have non-incremental processes are

7

more likely to travel through networks that carry information across state lines, through the

actions of policy entrepreneurs, organized interests and professional associations (Boushey 2010,

20-21). This could lead to reduced regional emulation. The percent of the population with

broadband subscriptions measures the inclusivity of technology use, indicating a generally richer

information environment in the states.

Digital Information, Economic Development and Society

Few transformations in history have been as rapid or as fundamental as the spread of

broadband internet. In less than 20 years, broadband has gone from a niche technology only used

by a few, to being found in over 70% of American households. If one includes cell phones to

connect to broadband networks, roughly 85% of Americans use the internet on a regular basis

(Pew Survey Research 2018). Broadband adoption in the United States results primarily from

private sector investments and market-driven practices (Crawford 2018, chapter 3) with very

small federal government investment. Approximately 500 communities have municipal

broadband or public-private partnerships (especially in rural areas where profitability is lower)

(Institute for Local Self-Reliance, accessed 4-10-19), but there is no state-provided broadband

anywhere in the nation.

Digital information now drives participation in politics, the economy, and society.

Longitudinal studies show that there is a wage premium for technology use at work and even at

home, controlling for other factors such as education (DiMaggio and Bonikowski 2008). This

wage premium exists not just for the most educated, but also for workers with a high school

education or less (Mossberger, Tolbert and McNeal 2008). Online political activities have also

been linked to voting and civic engagement for individuals (Tolbert and McNeal, 2003; see also

Boulianne 2009). Those who have broadband at home are more likely to acquire experience and

8

skill, and to engage in a range of economic and political activities online, including voting and

political news consumption (Mossberger et al 2008). Additionally, internet use fosters a range of

human-capital enhancing activities (DiMaggio et al. 2001; Hargittai 2002) by connecting

residents to information on jobs, transportation, health, online education, e-government (West

2004) and more. The internet has become the prevailing form of access to information, like

literacy and public education in previous centuries, and those without access are denied the

ability to effectively participate in society. In previous centuries literacy rates were a common

form of information access; we argue broadband subscriptions are a key measure of information

access in a digital era.

Information access and connectivity affects communities and states, not just individuals.

Place-based research on the effects of technology have primarily investigated its economic

benefits. Technology-fueled growth and the knowledge economy have increasingly concentrated

in a small number of superstar cities and states on both coasts with hot-spots in the middle of the

country (Berube and Murray 2018; Florida 2017), leading to growing inequality in wealth and

resources across the states or what Moretti (2012, 73) has called the “great divergence.” Favored

cities and states have a high presence of information technology and other knowledge-intensive

jobs, growing wages and growth of businesses establishments. Their regional economies are

based on ideas rather than production, with firms that cluster together to form complex

ecosystems of innovation (Moretti 2012, 9). Communities with more broadband access have

been able to leverage this technology to increase economic growth; might they also have the

resources to innovate in new policy ideas?

Research on broadband infrastructure has associated it with community benefits such as

faster job and firm growth (Lehr et al. 2006) and employment (Crandall, Lehr and Litan 2007),

9

decreased unemployment rates (Jayakar and Park 2013), knowledge intensive firms in non-

remote areas (Mack 2014), and other economic benefits (see Holt and Jamison 2009 for a

review). These studies, however, did not address the impact of broadband use by the

population. Beyond infrastructure, some new research finds broadband subscriptions are

powerful predictors of economic prosperity and wage growth for the nation’s 50 largest metros

over the past two decades, controlling for other known predictors of success (Mossberger,

Tolbert and Gracey 2018). Because broadband has spread unequally across the states (what is

known as the digital divide—Norris 2001; Mossberger et al 2003) and over time, there is

variation in digital information access.

Information and Policy Innovation

We expect states with richer digital information access will be more likely to adopt new

policies across the spectrum of issues over time. The spread of broadband internet has made the

flow of information much easier and faster than ever before. Broadband internet has changed

how information spreads from state to state by increasing the amount of information available,

and greatly reducing the cost of finding new information, from virtually anywhere. States have

always drawn upon multiple sources of information for policymaking, whether that is

professional networks, the media, think tanks, lobbyists, local and national parties, or myriad

other sources, both national and local (Mossberger 2000). The internet and digital media has

increased the ease of connecting not only policymakers, but advocates and citizens (see Boushey

2010). This leads us to a number of predictions. First, as broadband connectivity becomes more

prevalent in a state, states have access to substantially more information from citizens, interest

groups, and other states. This should result in states being more innovative in terms of adopting

new policies.

10

Hypothesis 1: States with higher levels of broadband adoption are more likely to adopt a policy

on average.

At the same time, broadband is expected to alter existing networks of information flows.

Broadband enables people to easily communicate across the country and globe. So, as the cost to

learn about what distant states are doing to tackle policy issues decreases, the role of geographic

proximity should decrease. Citizens can also be exposed to different state policy changes through

exposure on social media or other platforms and are more aware of what states across the country

are doing than ever before. Additionally, competition with other states has expanded from local

and regional, to the national, and even global scale. The internet has facilitated the concentration

of new industries in a select few tech hubs. As industry becomes more willing to relocate across

the country or world, localities must expand the scope of competition to a much larger set of

actors. Whether imitating, competing, or learning, states are expected to be less reliant on

contiguous actors than before. Contiguity will likely still be a factor in policy diffusion, but

broadband is expected to reduce its importance.

Walker’s (1969) seminal research proposed two mechanisms at play, although he didn’t

have empirical data at the time to test both. He saw geographic proximity with state policy

makers learning from their neighbors, but he also saw professional organizations and federalism,

with more national networks and information sharing. He argued that both were influential, but

that over time national professional networks might become more dominant. The literature has

focused on geographic proximity, while professional networks and information access have

received far less attention (Mossberger 2000; Balla 2001). Interest groups and policy

entrepreneurs also influence diffusion through national networks as well as within states

11

(Mintrom 1999). Broadband and digital media have not only made information access easier, it

has become more nationalized and less localized.

Hypothesis 2: Higher levels of broadband subscriptions over time should reduce the effect of

geography on policy diffusion.

Finally, we evaluate if broadband’s transformative potential applies to other predictors of

state policy diffusion. Large, wealthy, professionalized, urbanized and liberal states historically

been found to be the most innovative. Traditional predictors such as legislative professionalism

and education represent information access. So, too, does urbanization. Economist Ed Glaeser

(2011) has argued that innovation is more likely in dense urban areas because they increase the

flows of ideas and information. Today, the ease of digital information access may have allowed

other states to become more innovative. The low cost of searching for, or more passively,

receiving information may reduce the need for a highly professionalized legislature to learn

about new policy solutions, for example. Interest group networks have also grown considerably

in the digital era. States with above average broadband adoption may have different predictors of

policy innovation due to a richer, more networked and connected information environment. .

Hypothesis 3: Predictors of policy diffusion and innovation will be different in states with above

average broadband subscription rates compared to those with below average rates.

Research Design

We use the State Policy Innovation and Diffusion Dataset (SPID) collected by Boehmke

et al (2019) over the past two centuries. These data include information on thousands of policy

adoptions on hundreds of policies in all 50 states across a variety of policy areas. We use the

same approach as previous state policy diffusion research that uses a host of state political,

demographic, and economic variables to model policy innovation. We pair these data with data

12

on broadband subscriptions (adoption) in the states from 2000-2016 from the American

Community Survey and Current Population Survey from the Census Bureau. Figure 2 shows the

distribution of broadband subscription rates in the states over time. In the first time point, 2000,

no state has more than 10% broadband usage rates. Subscription rates quickly rise to between 25-

50% of the population having broadband subscriptions in 2005, with every state having over

60% subscriptions by 2015. While broadband has quickly spread across the entire country, there

are still large differences between the states, with gaps up to 25% between the states with the

highest and lowest rates of home broadband subscription. The combined data for 2000-2016

include over 1,600 adoptions of 105 policies, and nearly 29,000 total policy-state-year

observations. The dependent variable is a binary measure of whether a state adopted a policy in a

year.

To measure trends before broadband was widely available (the US Census began asking

questions on dial-up internet in 1997, and broadband in 2000) we compare the current time

period (2000-2016) to the last two decades of the twentieth century (1980-1999) as a control.

The 1980-1999 dataset includes more than 95,000 total policy-state-year observations. We also

create a pooled model from 1980-2016, where broadband subscriptions are coded 0 before the

year 2000. The combined dataset has nearly 170,000 observations and spans three and half

decades.

We use a pooled event history analysis (Kreizter and Boehmke 2016) to estimate the

probability of a state adopting a policy to account for internal and external predictors of state

policy innovation, with policies as a multilevel factor. A state becomes at risk of adopting a

policy after any state adopts a given policy (Berry and Berry 1990). So, if Hawaii adopts a

policy, the other 49 states become at risk of adopting that policy each year until they adopt the

13

policy. Once a state adopts a policy it drops out of the risk set. Each policy has its own risk set,

and an event history analysis models the probability of a state adopting a policy in a given year.

We follow Kreizter and Boehmke’s approach to pooling risk sets by including random intercepts

for each policy (2016). This method allows for pooling findings into a single, unified model

while also recognizing that each policy as a distinct baseline probability of adoption.

Our key independent variable is a measure of the proportion of households with a home

broadband subscription in each state over time. Our first model includes this variable in a typical

model of diffusion, with measures for previous contiguous state adoptions, as well as controls for

legislative professionalism, wealth (state income per capita), state population size, diversity

(percent black in the population), educational attainment, a measure of unified party control of

government, and state ideology (public opinion liberalism). We also include year fixed effects to

account for variations by year and a cubic polynomial for the amount of time that the state has

been at risk of adopting a policy. Finally, we use Desmarais et al’s (2015) measure of decayed

latent diffusion ties to measure trends over time, which is an inferred network using previous

policy adoptions that has been found to strongly predict policy adoption.

We first measure the effect of digital information access (broadband subscriptions) on

policy adoptions. The next set of models is the same specification, but with an interaction

between broadband and previous contiguous state adoptions. This will allow us to test if

broadband moderates the effect of contiguity on policy innovation. We then repeat this

interaction model but subsamples states by those with higher than average home broadband

subscriptions and those with lower than average subscriptions by year. With this approach, we

can test if being in more connected communities alters the relationship between standard

14

diffusion variables (such as legislative professionalism, state population, and wealth) and

innovativeness.

Finally, we estimate an instrumental variable model as a robustness check to demonstrate

the relationship between the growth of digital information and policy diffusion are not spurious.

We estimate a two stage probit that uses three geographic variables, state area, average elevation

in a state, and Dobson and Campbell’s (2014) measure of the percentage of a states area that is

flat. We believe argue these three variables affect broadband subscription rates but do not affect

policy innovation for a number of reasons (see appendix for more discussion of the instrumental

approach).

Simply put, people can only subscribe to broadband internet if the infrastructure exists.

Without access to broadband lines, people cannot sign up for a connection. Broadband lines are

expensive to install, so in larger states internet providers will have to devote more resources to

cover geographically larger states. Additionally, rugged terrain such as mountains will increase

the cost of laying lines for internet access, so providers will have an easier time installing

broadband infrastructure. When geographic barriers to building broadband infrastructure are low,

we expect broadband subscription rates to be higher. The second stage of the model estimates

policy adoption in the states, and uses fixed effects for year, a cubic polynomial for duration, and

fixed effects for policy to account for temporal dependence and difference baseline probabilities

of adoption between policies.

Results

The results in table 1 (model 1) show that both contiguity and latent network policy

adoptions predict policy innovation, consistent with previous research. States are more likely to

adopt a policy as the number of contiguous state adoptions increase, and as more states with

15

latent ties adopt that policy. Both findings match other research in diffusion that argue contiguity

and latent network ties increase policy innovation. The only significant controls are income per

capita and legislative professionalism, which negatively predict policy adoption and population

size which is positively signed. While singe policy studies have often found legislative

professionalism to positively predict innovation (Berry 1994, Shipan and Volden 2006) multiple

policy analyses (Boehmke and Skinner 2012) find legislative professionalism is negatively

associated with innovation. The key independent variable, digital information access as proxied

by broadband subscription, also increases the probability of a state adopting a policy. The

variables have been standardized so that their effect size can be compared, which shows that

broadband subscriptions has a larger effect than contiguity or latent network ties. These results

support our first hypothesis that broadband adoption increases innovativeness.

As a comparison, model 2 (table 1) replicates the analysis for the earlier time period

(1980-1999) as a control case omitting broadband. Without including a measure of digital

information access the effect size for latent network ties and geographic continuity are similar.

Model 3 pools data for the complete time period (1980-2016) confirming that broadband

subscriptions remain a powerful predictor of policy adoptions, controlling for standard predictors

of policy diffusion, including geographic continuity, latent ties, population size, liberal public

opinion and more educated populations. The effect size remains large.

Figure 3 shows the predicted probabilities for policy innovations at different levels of

broadband adoption. As the percentage of the population with broadband increases (across states

and over time), so does the probability of a state innovating. The effect is substantively very

large. The probability of innovation more than doubles from less than a 5% probability of

adoption to over a 10% probability as a state’s population goes from low to high levels of

16

broadband subscriptions. A roughly 10% increase in broadband subscription results in a 1%

increase the probability of adoption. An increase of 1% is substantively very large when

considering that the baseline probability of adopting a policy in any given year is just under 5%

(the red line in figure 3). So, a 10% increase in broadband subscription rates increases the

baseline probability of adoption by roughly 20%. Broadband adoption has a positive and

substantively large relationship with policy adoption.

The results in table 2 show the interaction between contiguity and broadband

subscriptions. As in the previous table, both broadband subscriptions and contiguous adoptions

increase the probability of adoption (when ignoring the interaction). However, the interaction

between the two is negative and statistically significant across the three model specifications—1)

current time period column 1, 2) pooled time period column2, and 3) pooled time period with

additional binary variable. Model 3 adds a binary variable for years before 2000 as an additional

robustness check to isolate variation between the two eras. This means that increases in

broadband subscriptions decrease the effect of geographic continuity in states innovating. Figure

4 shows the marginal effect of contiguity by broadband adoption rate (model 1, table 2). The

effect of contiguity is positive, significant, and contiguous state adoptions increase the

probability of adoption by roughly three-fourths of a percentage point with each additional

contiguous adoption. However, the marginal effect of contiguity decreases as broadband

subscription rates surpass 60% of the population. Contiguity no longer positively predicts policy

adoption when subscription rates are over 85%. These results support our second hypothesis that

broadband has altered flows of information and reduced the role of geography in policy

diffusion. The states with high levels of digital information appear to no longer rely on contiguity

for policies.

17

Table 3 compares states with above average and below average broadband subscription

rates for the current time period (2000-2016). We find that there are significant differences in

high and low broadband states. The base term for contiguity is significant for both models, but

broadband subscriptions only significantly predict policy adoption in states with above average

subscription rates. We find that broadband subscription rates reduce the effect of contiguity in

high broadband states, but not in low broadband states. Figure 5 shows the marginal effect of

contiguity on policy adoption in high broadband states. Each contiguous state adoption increases

the probability of a state innovating by roughly 1% when broadband subscription rates are below

60%. Contiguity’s effect weakens as broadband subscriptions increase, and the effect disappears

with subscription rates above 80%. Contiguity does not significantly predict policy adoption in

states with high rates of broadband usage.

We also find differences in several other variables in the model. Latent ties only predict

policy adoption in below average broadband states, as does population. Legislative

professionalism lowers the probability of states innovating in low broadband states but has no

effect in high broadband states. Income decreases innovation in high broadband states, but not in

below average broadband states. These results support our third hypothesis that broadband

adoption alters the political context of a state which in turn has effects on the diffusion network.

Additionally, we find more supporting evidence that broadband adoption reduces the role of

geography in policy diffusion.

Finally, table 4 shows the results for the two-staged probit model that treats broadband

subscription rates as an endogenous predictor of policy adoption. In the first stage of the model,

larger states have lower levels of broadband subscription rates, and flatter states on average have

higher rates of broadband subscription. States with an average higher elevation have somewhat

18

higher rates of broadband subscription. The other demographic variables behave as expected.

Wealthy, well educated, and populous states have higher levels off broadband adoption.

The second stage models policy adoption. Broadband subscriptions are a large,

positively, and statistically significant predictor of policy adoption in the states. The predicted

probability of adoption ranges from 4-percent for states with low levels of broadband

subscriptions (below 20 percent) up to 7.6-percent for states with near universal levels of

broadband (over 90 percent). The three percentage point increases in the probability of adoption

is substantively low when considering the overall baseline probability of adoption is very low.

Many of the other variables in the dataset behave as expected. Adoptions from contiguous states

or states with latent ties increase the probability of adoption, and more populous states tend to be

more innovative. Income per capita predicts higher levels of broadband subscription but not

policy adoption, which provides further support for our argument that the relationship between

broadband subscriptions and policy adoption is not spurious. Through a variety of specifications

we have shown that broadband subscriptions strongly and positively predict policy

innovativeness.

Discussion and Conclusion

Since Walker’s (1969) work, diffusion scholars have cited the central role of information

in the spread of policies between states. Whether through imitation, learning, or competition,

states must be aware of what other states are doing in order to adopt their policies. Learning and

competition require even more information as states must be aware not only of what policies are

being adopted, but the policies’ effects. As diffusion studies have become increasingly

sophisticated, scholars have identified several predictors that go beyond contiguity to explain the

spread of policies, including latent network ties (Desmarais et al 2015) and interest group

19

activity (Garrett and Jansa 2015, Gandara, Rippner, and Ness 2017), while others have identified

that contiguity is playing a less prominent role in diffusion over the past few decades (Mallinson

2019). We argue that the rapid spread of the internet has altered traditional diffusion pathways

because it has dramatically increased the amount of information available while decreasing the

cost to access information. Broadband adoption facilitates rapid and low-cost communication

across the world between interest groups, the citizenry, and states.

The research presented here supports our argument that broadband subscription rates are

positively associated with policy innovation. We believe that broadband increases innovation

through a number of mechanisms. Following prior research in policy diffusion, these include

bureaucratic processes drawing upon professional networks (Walker 1969; Mossberger 2000;

Balla 2001), nationally-connected interest groups and policy entrepreneurs (Mintrom 2000;

Kingdon 1995; Schneider and Teske 1995), groundswells of public support (Boushey 2010;

Pacheco 2012). We demonstrate that digital information increases innovativeness, and that its

effect is substantively large, with a 10% increase in subscription rates increasing the baseline

probability of adopting a policy in a given year by roughly 20%. This relationship holds under

other specifications including an instrumental variable approach that controls for factors that may

cause both policy innovation and broadband subscriptions.

We find that broadband adoption not only increases innovativeness in the states, but also

alters the flow of policies. Communities with high broadband usage can easily look beyond their

immediate neighbors for policy solutions, and citizens and interest groups are increasingly able

to operate at a national level. Broadband internet has facilitated this change and altered the

pathways by which policies traditionally diffuse. States with high levels of digital information

are less reliant on contiguity for policy or other traditional sources of information, such as latent

20

networks or legislative professionalism. We also evaluate how broadband context affects policy

diffusion. We find differing predictors for states with above average and below average

broadband adoption. States with higher rates of broadband usage tend to be less reliant on

geography, latent networks, and population when innovating. We argue this reflects that

communities with high broadband usage are more interconnected nationally, and may also have a

more active citizenry and robust interest group population. Broadband internet is a resource to

easily identify new policy solutions through learning, imitation, and competition.

Diffusion scholars should incorporate the role of broadband internet in models of

diffusion because it represents an important way in which information can flow between the

states. This is just the first step in understanding how digital information affects policy diffusion.

Future studies could evaluate whether broadband has a larger effect on the more information

intensive diffusion mechanisms of learning and competition compared to imitation. Additionally,

digital information could have policy domain specific effects. Is diffusion more likely for

policies with greater political salience and public attention, for example, where a large

proportion of the population has broadband? Or, are national influences more likely for policies

with greater risk and uncertainty, as policymakers may be more motivated to engage in

information search? These findings across thousands of policies, however, add to our confidence

that digital information is a neglected aspect of the policy diffusion literature worthy of further

exploration.

21

Bibliography

Balla, S. J. 2001. Interstate professional associations and the diffusion of policy Innovations. American Politics Research 39:221–45.

Baumgartner, F. R. and B. D. Jones. 1993. Agendas and instability in American politics. University of Chicago Press.

Berry, F. S., & Berry, W. D. (1990). State lottery adoptions as policy innovations: An event history analysis. American political science review, 84(2), 395-415.

Berry, F. S. (1994). Sizing up state policy innovation research. Policy Studies Journal, 22(3), 442-456.Berry, W. D., and Baybeck, B. (2005). Using geographic information systems to study

interstate competition. American Political Science Review 99(4): 505–19.Berube, A., & Murray, C. (2018). Renewing America's Economic Promise Through Older Industrial

Cities. Metropolitan Policy Program at Brookings.Boehmke, F. J., & Skinner, P. (2012). State policy innovativeness revisited. State Politics & Policy

Quarterly, 12(3), 303-329.Boushey, G. (2010). Policy diffusion dynamics in America. Cambridge University Press.Boulianne, S. (2009). Does Internet use affect engagement? A meta-analysis of research. Political

communication, 26(2), 193-211.Chadwick, A.. (2017). The hybrid media system: Politics and power. Oxford University Press.Crandall, R. W., Lehr, W., & Litan, R. E. (2007). The effects of broadband deployment on output and

employment: A cross-sectional analysis of US data.Crawford, S. (2018). Fiber: The coming tech revolution – and why America might miss it. New

Haven, CT: Yale University Press.Desmarais, B. A., Harden, J. J., & Boehmke, F. J. (2015). Persistent policy pathways: Inferring diffusion

networks in the American states. American Political Science Review, 109(2), 392-406.DiMaggio, P., Hargittai, E., Neuman, W. R., & Robinson, J. P. (2001). Social implications of the

Internet. Annual review of sociology, 27(1), 307-336.DiMaggio, P., & Bonikowski, B. (2008). Make money surfing the web? The impact of Internet use on the

earnings of US workers. American Sociological Review, 73(2), 227-250.Dobson, J. E., & Campbell, J. S. (2014). The flatness of US states. Geographical Review, 104(1), 1-9.fFlorida, R. (2017). The new urban crisis: How our cities are increasing inequality, deepening

segregation, and failing the middle class—And what we can do about it. Basic Books.Gandara, D., Rippner, J. A., & Ness, E. C. (2017). Exploring the ‘how’ in policy diffusion: National

intermediary organizations’ roles in facilitating the spread of performance-based funding policies in the states. The Journal of Higher Education, 88(5), 701-725.

Garrett, K. N., & Jansa, J. M. (2015). Interest group influence in policy diffusion networks. State Politics & Policy Quarterly, 15(3), 387-417.

Gilardi, F. (2016). Four ways we can improve policy diffusion research. State Politics & Policy Quarterly, 16(1), 8-21.

Glaeser, E. (2011) The Triumph of the City. New York: Penguin Press. Gray, V. (1973) “Innovation in the States: A Diffusion Study.” American Political Science

Review 67 (4): 1174–185.Grossback, L. J., Nicholson-Crotty, S., & Peterson, D. A. (2004). Ideology and learning in policy

diffusion. American Politics Research, 32(5), 521-545.Hargittai, E. (2002). Beyond logs and surveys: In‐depth measures of people's web use skills. Journal of

the American Society for Information Science and technology, 53(14), 1239-1244.Hopkins, D. J. (2018). The increasingly united states: How and why American political behavior

nationalized. University of Chicago Press.Jayakar, K., & Park, E. A. (2013). Broadband availability and employment: An analysis of county-level

data from the National Broadband Map. Journal of Information Policy, 3, 181-200.

22

Karch, A., Nicholson-Crotty, S. C., Woods, N. D., & Bowman, A. O. M. (2016). Policy diffusion and the pro-innovation bias. Political Research Quarterly, 69(1), 83-95.

Kreizter, R. J., & Boehmke, F. J. (2016). Modeling heterogeneity in pooled event history analysis. State Politics & Policy Quarterly, 16(1), 121-141.

Lehr, W., Sirbu, M., & Gillett, S. (2006). Wireless is changing the policy calculus for municipal broadband. Government Information Quarterly, 23(3-4), 435-453.

Lowi, T. (1972) Four systems of policy, politics, and choice " Public Administration Review 298-310.Mack, E. A. (2014). Broadband and knowledge intensive firm clusters: Essential link or auxiliary

connection?. Papers in Regional Science, 93(1), 3-29.Mallinson, D. J. (2019). Who Are Your Neighbors? The Role of Ideology and Decline of Geographic

Proximity in the Diffusion of Policy Innovations. Policy Studies Journal.Mintrom, M. (2000.) Policy entrepreneurs and school choice. Georgetown University Press. Mintrom, M., and S. Vergari. 1998. Policy networks and innovation diffusion: The

case of education reform.” Journal of Politics 60:126–48.Moretti, E. 2012. The new geography of jobs. New York: Mariner Books. Mossberger, K. (2000) The Politics of Ideas and the Spread of Enterprise Zones. Georgetown University

Press.Mossberger, K. and K. Hale. 2002. Poly diffusion in intergovernmental programs: Information diffusion

in the school-to-work network. The American Review of Public Administration.Mossberger, K., C. Tolbert and R. McNeal. 2008. Digital citizenship: The Internet, society and

participation. MIT Press.Mossberger, K., C. Tolbert and W. Franko. 2012. Digital cities: The Internet and the geography of

opportunity. Oxford University Press.Nicholson-Crotty, S., & Carley, S. (2018). Information exchange and policy adoption decisions in the

context of US state energy policy. State Politics & Policy Quarterly, 18(2), 122-147.Norris, P.. (2001). Digital divide: Civic engagement, information poverty and the internet worldwide.

Cambridge University Press.Pacheco, J. (2012). The social contagion model: Exploring the role of public opinion on the diffusion of

antismoking legislation across the American states. The Journal of Politics, 74(1), 187-202.Rogers, E.  1962.  The diffusion of innovations.  NY:  Free Press.Rogers, E.  2003.  The diffusion of innovations, 5th ed.  NY:  Free Press.Rose, R. 1993. Lesson-drawing in public policy: A guide to learning across time and space. Chatham

House. Shipan, C. R., & Volden, C. (2006). Bottom‐up Federalism: the diffusion of antismoking policies from

US cities to states. American journal of political science, 50(4), 825-843.Shipan, C. R., & Volden, C. (2008). The mechanisms of policy diffusion. American journal of political

science, 52(4), 840-857.Silver, Nate. The signal and the noise: The art and science of prediction. Penguin.Sovey, A. J., & Green, D. P. (2011). Instrumental variables estimation in political science: A readers’

guide. American Journal of Political Science, 55(1), 188-200.Walker, J. L. (1969). The diffusion of innovations among the American states. American

Political Science Review 63 (3): 880–99.West, D. M. (2004). E‐government and the transformation of service delivery and citizen attitudes. Public

administration review, 64(1), 15-27.

23

FiguresFigure 1. Percent of US Adults who are Home Broadband Users

Figure 2: Histogram of Broadband Subscription Rates in the States (2000, 2005, 2010, and 2015)

24

Table 1: Pooled Event History Analysis of State Policy Innovations (2000-2016)*(1) (2) (3)

2000-2016 1980-1999 Pooled 1980-2016

Broadband Subscriptions 1.7987** 0.7553**(0.6044) (0.3599)

Latent Ties 0.0843*** 0.1012*** 0.1329***(0.0219) (0.0112) (0.0084)

Geographic Contiguity 0.1584*** 0.1893*** 0.1776***(0.0272) (0.0177) (0.0118)

Population 0.0542 0.1239*** 0.1000***(0.0355) (0.0262) (0.0180)

Public Opinion Liberalism 0.0219 0.1352*** 0.1323***(0.0340) (0.0329) (0.0193)

Unified Party Control Govt 0.0433 -0.0820** -0.0146(0.0552) (0.0353) (0.0253)

Income Per Capita -0.1111** -0.0732* -0.0686**(0.0426) (0.0407) (0.0240)

Legislative Professionalism -0.0877* -0.0958*** -0.1033***(0.0453) (0.0288) (0.0213)

Prop Black -0.0080 0.0088** 0.0080***(0.0155) (0.0032) (0.0024)

Percent High School 0.0119 0.0222*** 0.0173***(0.0109) (0.0048) (0.0035)

Duration -0.0250 -0.1975*** -0.0425**(0.0562) (0.0304) (0.0138)

Duration Squared -0.0022 0.0294*** 0.0017(0.0112) (0.0045) (0.0012)

Duration Cubed 0.0001 -0.0012*** -0.0000(0.0006) (0.0002) (0.0000)

Constant -4.7767*** -3.6447*** -3.3404***(1.0183) (0.4267) (0.3147)

var(policy) 0.8401*** 1.1791*** 0.9505***(0.1328) (0.1217) (0.0792)

Observations 28850 95346 170363*indicates p<.05. Modeling includes fixed effect for year and random effects for policy. More policies began diffusing in the 80s and 90s than those that began diffusing the in the 2000s, so the number of observations is larger for the earlier period. Many of the policies from the 2000s are still diffusing, while most from the 80s and 90s fully diffused. Broadband home adoption first measured by the Census Current Population Survey in 2000. Internet use dial-up first measured by CPS in 1997.

25

Figure 3: Predicted Probability of Adopting any given Policy in a Year (model 1, table 1)

Note: Predicted probabilities are generated from the fixed component of model 1 (Table 1) and reflect the averaged random effects. The red line is the baseline probability of adoption of any policy for the entire sample (4.8%). Shaded area represents 95% confidence interval around the probability estimate.

26

Table 2: Pooled Event History Analysis of State Policy Innovations with Interaction between Geographic Contiguity and Broadband Subscriptions *

(1) (2) (3)2000-2016 Pooled 1980-

2016Pooled- Pre

2000 DummyBroadband Subscriptions 1.9378** 0.7390** 0.7073*

(0.6079) (0.3596) (0.3619)Geographic Contiguity 0.3943*** 0.2215*** 0.2223***

(0.0996) (0.0148) (0.0152)Broadband Subscriptions # Geographic Contiguity

-0.3570** -0.1745*** -0.1681***

(0.1458) (0.0364) (0.0367)Latent Ties 0.0764*** 0.1287*** 0.1339***

(0.0222) (0.0085) (0.0087)Population 0.0549 0.1000*** 0.1003***

(0.0355) (0.0180) (0.0182)Public Opinion Liberalism 0.0176 0.1312*** 0.1270***

(0.0340) (0.0193) (0.0196)Unified Party Control Govt 0.0448 -0.0157 -0.0105

(0.0552) (0.0254) (0.0258)Income Per Capita -0.1129** -0.0680** -0.0732**

(0.0426) (0.0240) (0.0246)Legislative Professionalism -0.0855* -0.1020*** -0.0999***

(0.0453) (0.0213) (0.0216)Prop Black -0.0087 0.0086*** 0.0094***

(0.0155) (0.0024) (0.0025)Percent High School 0.0114 0.0183*** 0.0198***

(0.0109) (0.0035) (0.0035)Duration -0.0305 -0.0442** -0.0508***

(0.0564) (0.0138) (0.0141)Duration Squared -0.0001 0.0020 0.0026**

(0.0112) (0.0012) (0.0013)Duration Cubed -0.0000 -0.0000 -0.0000

(0.0006) (0.0000) (0.0000)Before 2000 -0.1432

(0.4730)Constant -4.5430*** -3.3239*** -3.2978***

(1.0255) (0.3148) (0.3543)var(policy) 0.8497*** 0.9581*** 0.9422***

(0.1342) (0.0799) (0.0791)Observations 28850 170363 163344*indicates p<.05. Modeling includes fixed effect for year and random effects for policy

27

Figure 4 Marginal Effect of Contiguous Adoptions by Broadband Subscription Rate on Policy Adoption (model 1, table 2)

28

Table 3: Pooled Event History Analysis of State Policy Innovations (2000-2016) Sub-sampled by above Average and below Average Broadband Subscription Rates*

(1) (2)High Low

Broadband Subscriptions 1.9481** 1.7292(0.8237) (1.4224)

Geographic Contiguity 0.6787** 0.4337**(0.2292) (0.1734)

Broadband Subscriptions # Geographic Contiguity

-0.7232** -0.6251

(0.3133) (0.4527)Latent Ties 0.0422 0.0923**

(0.0284) (0.0376)Population -0.0107 0.2426***

(0.0428) (0.0728)Public Opinion Liberalism 0.0220 0.0208

(0.0428) (0.0659)Unified Party Control Govt 0.0206 0.1018

(0.0679) (0.1013)Income Per Capita -0.1426** 0.0080

(0.0490) (0.1014)Legislative Professionalism -0.0360 -0.2289**

(0.0522) (0.0959)Prop Black 0.0982 -0.0124

(0.4629) (0.0163)Percent High School 0.0170 0.0165

(0.0296) (0.0149)Duration -0.0619 0.2728*

(0.0687) (0.1626)Duration Squared 0.0017 -0.1212**

(0.0135) (0.0597)Duration Cubed 0.0001 0.0110*

(0.0007) (0.0060)Constant -4.6035* -4.9436***

(2.6409) (1.3353)Var(policy) 1.0831*** 0.6663***

(0.1823) (0.1790)Observations 18971 9879

*indicates p<.05. Modeling includes fixed effect for year and random effects for policy

29

Figure 5: Marginal Effect of Contiguity in Above Average Broadband States (model 1, table 3)

Figure 6 Adoptions Per Year*

*State Policy Innovation and Diffusion Database (Boehmke et al 2019)

30

Table 4: 2 Stage-Probit with Broadband Treated as Instrumental VariableStage 1- Stage 2-

Broadband Subscriptions

Policy Adoption

Broadband 1.1958***(0.3112)

Contiguity -.0009 0.0682***(.0011) (0.0148)

Latent Decay -.0014 0.0340**(.0010) (0.0122)

Population .0129*** 0.0330*(.0015) (0.0180)

Citizen Ideology .0546*** -0.0394*(.0011) (0.0229)

Unified control .0324*** -0.0067(.0019) (0.0304)

Real Income Per Capita .1215*** -0.1801***(.0013) (0.0433)

Squire Professionalism -.0158*** -0.0284(.00156) (0.0232)

Percent High School .0077*** 0.0005(.0003) (0.0064)

Duration -.0035 -0.0008(.0026) (0.0362)

Duration Squared .0002 -0.0029(.0003) (0.0055)

Duration Cubed .000008 0.0001(.00002) (0.0003)

Geographic Area -.0602***(.0034)

Flatness .0016***(.0001)

Average Elevation .00003***(.0000007)

Constant -2.568004 0.1936(.0349) (1.0043)

Observations 27689Wald test of exogeneity: χ2 11.37 (p<.001)

*fixed effects for year and policy included

31