Twitter Adoption and Activity in U.S. Legislatures: A 50-State Study

Abstract

This study draws inspiration from the literature on Twitter adoption and activity in U.S. legislatures, applying predictions from those limited studies to all 7,378 politicians serving across 50 American state legislatures in the fall of 2015. Tests of bivariate association carried out for individual states lead to widely varying results, indicating an underlying diversity of legislative environments. However, a pooled multivariate analysis for all 50 states indicates that constituents per legislator, the youth and educational attainment of a district, legislative professionalism, being a woman, sitting in the upper chamber, leadership, and legislative inexperience are significantly and positively associated with Twitter adoption and current Twitter use. Controlling for these factors, neither legislator party, nor majority status, nor partisan instability, nor district income is significantly related to either Twitter adoption or current Twitter use. Although women are more likely than men to adopt and use Twitter, the most active users narrowly tend to be men. Finally, most variation in social media adoption and activity by legislators remains unexplained, leaving considerable room for further theoretical development.

Keywords

social media political communication Twitter

Introduction

With the short-form social media platform Twitter past its 10th birthday, its place in American national politics seems secure. Less than a week after winning office, President Donald Trump openly credited Twitter for his victory:

It’s a great form of communication . . . I’m not saying I love it, but it does get the word out. When you give me a bad story or when you give me an inaccurate story . . . I have a method of fighting back . . . I think it helped me win all of these races where they’re spending much more money than I spent. And I won. I think that social media has more power than the money they spent, and I think maybe to a certain extent, I proved that. (Stahl, 2016)

Combating traditional media narratives through independent dissemination of information is only one use of Twitter by politicians, but sharing of information (mostly expressing legislators’ positions) appears to be the primary purpose in over half of U.S. Congressional Twitter posts. Noting an activity or a location in a home district is the purpose of roughly a quarter of these congressional “tweets,” while personal messages, requests for action and fundraising are quite rare (Glassman, Straus, & Shogan, 2013; Golbeck, Grimes & Rogers, 2010). For these legislators, Twitter is a powerful new direct broadcast medium.

As indicated above, studies of Twitter use by U.S. legislators tend to focus on the U.S. Congress, where 96.6% of Representatives and 100% of Senators have obtained accounts on Twitter (Shpayer, 2016). In Congress, adoption appears to be a constant rather than a variable. At the subnational level, however, the question of Twitter adoption by state legislators is less settled. Although much important legislation passes through state legislatures, few studies of social media at the state level have been conducted. One aim of this study is to fill that gap. Have U.S. state legislators adopted Twitter with the same ubiquity as their federal counterparts? Do those legislators with Twitter accounts use the service actively? How actively?

This study finds that Twitter adoption is far more variable at the state legislative level. Given this, a second aim of this study is to understand why some state legislators take to Twitter, while others do not. Do theoretical claims from the limited and aging literature on early legislative Twitter use explain the behavior of the 7,378 American state legislators in recent times? The large number of state legislators, organized into 50 exhaustive, mutually exclusive, and independently operated bodies, allows in a sense for a 50-fold replication of earlier work. Results reveal that rates of Twitter adoption and active Twitter use are highly variable, with their correlates varying considerably from state to state.

A 50-state study is not only useful for purposes of replication but also because it throws larger scale structural features of political context into comparative relief. In addition to traditionally studied district-level and individual-level effects, state-level effects can also be assessed in a pooled model. When legislators are pooled, many individual, district, and contextual variables are associated with social media adoption and use, but much variation remains unexplained. This article concludes with a consideration of implications for theory and research.

Literature and Theory

Despite the vital role of legislatures in American politics, the existing literature on Twitter adoption and activity is slight, focused primarily on the U.S. Congress, and further focused on the period of 2009-2010 when Twitter use by members of Congress was no longer an oddity but not yet universally expected (Chi & Yang, 2010; Glassman, Straus, & Shogan, 2009; Golbeck, Grimes & Rogers, 2010; Gulati & Williams, 2010; Lassen & Brown, 2010; Peterson, 2012). Exceptions to this rule are two studies of Twitter adoption in 2012 (Evans, Cordova, & Sipole, 2014; Glassman et al., 2013) and three studies of legislatures in Texas (Sala & Jones, 2012), Wisconsin (Moran, 2013), and New England (Williams, Gulati, & DeLeo, 2013). Across these studies, three classes of theoretical explanations for Twitter adoption emerge, focusing on characteristics of individual legislators, legislative districts, and the political environment.

Characteristics of Legislators

Especially in the United States, studies of political behavior have focused on the strategic decision making of individual politicians. It is frequently asserted (Chi & Yang, 2010; Evans et al., 2014; Lassen & Brown, 2010; Moran, 2013; Peterson, 2012; Williams et al., 2013) that legislature neophytes will be more likely to adopt Twitter than veteran legislators. Electorally vulnerable and more resource-poor, new legislators may find Twitter an inexpensive means for building a political brand (Lassen & Brown, 2010; Peterson, 2012). This pattern is borne out in the U.S. Congress of 2012 (Evans et al., 2014), but otherwise, no association between legislative experience and Twitter adoption has been found.

Legislators who hold leadership positions are committed to their legislative identity, but an inward focus on relations with other legislators may render them less attentive to external communication (Marmaduke, 2015). As power brokers with access to traditional modes of influence, leaders may rely less on innovative modes of influence. However, the sole study to include legislative leadership as an independent variable finds that it is unassociated with Twitter adoption (Chi & Yang, 2010).

Finally, it has been hypothesized that women legislators in the United States will be more likely to adopt and use Twitter, simply because women in the general U.S. population have been disproportionate users of Twitter in the past (Chi & Yang, 2010; Evans et al., 2014; Gainous & Wagner, 2014; Moran, 2013), although the gap has substantially narrowed in recent years (Greenwood, Perrin, & Duggan 2016). More junior in legislative experience and less likely to hold leadership positions in nonfeminized policy interest areas (Reingold & Smith, 2014), women legislators may have a greater incentive to use newer, more accessible levers of power like Twitter. Empirical evidence is mixed; while some studies find that women are more likely to adopt Twitter in legislatures (Evans et al., 2014; Moran, 2013), other research finds no such association (Chi & Yang, 2010).¹

Characteristics of Districts

If legislators are strategic in their behavior, then legislators who represent districts heavy in Twitter use will be more inclined to adopt Twitter themselves to reach constituents (Gulati & Williams, 2010; Lassen & Brown, 2010; Peterson, 2012). Americans who are younger, with higher income and higher education are more likely to use Twitter (Greenwood et al., 2016). Therefore, legislators in districts with more young, high-income, highly educated constituents may be more likely to adopt and use Twitter.

Empirical support for this prediction is mixed. On the dimension of education, one study finds the predicted effect in the U.S. House but not the U.S. Senate (Lassen & Brown, 2010); other studies find no effect in either chamber (Gulati & Williams, 2010; Moran, 2013; Peterson, 2012). On the dimension of district age, one congressional study concluded that representatives of older districts are less likely to adopt Twitter (Gulati & Williams, 2010), but another found no such association (Peterson, 2012), and another found empirical support varying by chamber (Lassen & Brown, 2010). Representatives of high-income districts in Wisconsin are more likely to adopt Twitter (Peterson, 2012), but in the U.S. House of 2010, representatives of high-income districts were actually less likely to use Twitter (Lassen & Brown, 2010). Other studies of the Congress in 2010 find no effect of district income (Chi & Yang, 2010; Peterson, 2012).

Characteristics of Political Environment

Observing in that in 2009 and 2010, Republicans members of Congress were disproportionate users of Twitter (Chi & Yang, 2010; Glassman et al., 2009; Golbeck, Grimes, & Rogers, 2010; Gulati & Williams, 2010; Lassen & Brown, 2010; Peterson, 2012), scholars made general explanations that were largely based in theory and not subjected to empirical test. Some scholars suggest Twitter is attractive to Republicans because it circumvents mainstream news media that Republican politicians consider biased (Gainous & Wagner, 2014). Perhaps the Republican wave of congressional Twitter use was due to masterful steering by insightful congressional leadership (Lassen & Brown, 2010). It could also be that the Republican Party more durably embodies a coherent political brand, and that Twitter is particularly amenable to succinct branding efforts (Gainous & Wagner, 2014).

Regardless of Republicans’ early dominance of the platform, evidence for an enduring Republican advantage on Twitter is weak (Evans et al., 2014). Studies of state legislatures in Texas and New England found no effect of party on Twitter adoption (Sala & Jones, 2012; Williams et al., 2013), and in Wisconsin, Democratic state legislators were actually more likely to adopt Twitter (Moran, 2013). This pattern could be explained by a party minority effect. Minority parties cannot shape the legislative agenda and lose control of the leadership positions that shape the public messages of a legislature (Evans et al., 2014; Lassen & Brown, 2010; Moran, 2013). While majority parties are more complacent, minority parties have a stronger incentive to innovate as a way of regaining power (Lowi, 1963; Peterson, 2012).

Finally, upper-chamber senators in the U.S. Congress have a slight but consistent higher rate of Twitter adoption than lower chamber representatives (Glassman et al., 2009, 2013). General theoretical accounts explaining this difference have not been developed, but upper chambers of legislatures tend to have longer tenures, making the investment in learning Twitter more worthwhile. Upper-chamber legislators have larger districts and many more constituents; Twitter may help senators communicate effectively at scale while maintaining a folksy feel. Senators in the U.S. Congress and many states have more staff support than representatives do, and those staffers may help create and maintain a Twitter account.

Characteristics of States

Because research to date has not covered all state legislatures, it has not been possible to generate and test comparative hypotheses. Because this study observes 50 independent state legislative environments, comparative hypotheses invoking variation between states are possible, allowing a preliminary glimpse at structural factors that may influence Twitter adoption.

First, Twitter is an Internet-based application, and access to the Internet varies significantly across the United States. According to American Community Survey data, the percentage of households with some form of Internet subscription varies from a low of 59.9% for the state of Mississippi to a high of 82.7% for the state of New Hampshire (U.S. Census Bureau, 2009-2014). In states with higher Internet connection rates by constituents, it is reasonable to expect greater Twitter adoption rates among state legislators.

Second, while the number of legislators in American state legislatures varies by a factor of slightly less than 10 (from Nebraska’s 49 legislators to New Hampshire’s 424), the number of constituents per legislator varies by a factor of slightly more than a hundred (from New Hampshire’s 3,116 constituents per legislator to California’s 317,224 constituents per legislator). Residents of states with few constituents per legislator can expect more personal access to their representatives; members of the Maine State Legislature (7,181 constituents per legislator) regularly publish their home addresses and phone numbers. In such states, dense offline social networks may obviate the need for active online social networking. Politicians in states with many constituents per legislator cannot maintain the same level of personal offline access, and may therefore take to Twitter for easily scalable communications.

Third, the professionalism of a legislature may affect the rate of Twitter adoption. Measured in the Squire Index as a combination of compensation, expectations of time investment, staff, and resources (Squire, 2007), greater professionalism creates a greater opportunity for legislators to invest in representative efforts such as setting up and using a Twitter account, or assigning staff to do so on legislators’ behalf.

Fourth, if we take the words of Theodore Lowi to heart, “consolidation is the opposite of innovation” (Lowi, 1963). If Lowi is right, stable arrangements of legislative power can be expected to discourage innovation while disruptions of political stability through periodic shifts in partisan control can be expected to promote innovation. To the extent that adoption of Twitter represents an innovation, we can expect that states with partisan shifts or splits will have higher rates of Twitter adoption. For a strong contrary position, see Mayhew (2000, 2005), who notes the plausibility of Lowi’s theory but questions the empirical basis for such claims.

Data and Method

Data are gathered from a variety of sources. Information regarding the names, partisan affiliations, chambers, length of service, and district assignments of members of state legislatures as of September 2015 was gathered from Open States, a government transparency project of the Sunlight Foundation (2015). A random selection of Open States data was checked against state legislative websites for currency and accuracy. Nebraska legislators were dropped from analyses involving chamber and party data, since Nebraska’s legislature is both unicameral and nonpartisan. Representative gender was measured through a combination of name, profile photo, and pronoun use in representatives’ official legislative biographies and Twitter feeds. Leadership lists, partisan balance, and chamber hierarchy were obtained from each state legislature’s website.

Variables describing characteristics of states and state legislative districts were obtained from American Community Survey 2009-2014 5-year summary files using current state legislative boundaries. An exception regards 52 of the 424 legislative districts of New Hampshire, data for which were not available from the American Community Survey; bivariate analyses were run for these districts but they were dropped from multivariate analysis due to missing district-level data. District-level ACS-derived variables include percent population with bachelor’s degree or higher, population per representative, median income, and median age. Percentage of households with Internet access is obtained from the ACS at the state level because as a new variable, it is unfortunately not yet available at the state legislative district level. An additional state-level variable is “Party Switch/Split,” a dichotomous measure assessing whether a state legislature switched party control or experienced split party control during the years 2010-2015, according to the National Conference of State Legislatures (Warnock, 2015). Finally, the Squire Index of professionalism for each state was included, with higher values indicating greater professionalism (Squire, 2007).

Information about the adoption and use of Twitter accounts by legislators was obtained through a multifaceted search strategy. Existing Twitter lists of state politicians, Twitter searches, Google searches, legislators’ official and campaign web pages were an initial source of potential state legislator Twitter accounts. Each of these accounts was checked to rule out selection errors and parody accounts. In a truncated snowball sample, a list of accounts following and being followed by this initial pool of legislators’ accounts was generated and checked for additional state legislators. The result is a list of legislative Twitter accounts for each state. Once the lists were assembled, each Twitter account on the list was accessed again to determine the number of “tweets” made by each account and whether any posts by an account had been made in 2014 or 2015.

After obtaining univariate statistics, bivariate analyses were run describing the associations of nine individual- and district-level independent variables with the dichotomous dependent variables Twitter adoption (1: Adopted), Twitter posting in 2014-2015 (1: Posted) and large-volume posting by those active in 2014-2015 (1: ≥500 tweets). These three dependent variables reflect increasingly strong thresholds of Twitter use and activity. The first two of these variables were measured for all legislators; the third variable was only measured for those already known to have active Twitter accounts in 2014-2015, leading to a smaller number of cases. The goal of these bivariate tests was to depict the variety of findings researchers would obtain if they only studied a single state legislature. The direction of association between variables was noted, p values were obtained (on the basis of chi-square tests for dichotomous independent variables and t tests for continuous independent variables), and combinations of direction and p value were counted for each of the 50 states (due to its unicameral nonpartisan legislature Nebraska was omitted for chamber and party variables).

Multivariate logistic regressions were run that pooled legislator data from all 50 states for the same independent variables and dependent variables used for bivariate tests. A second set of models added the four state-level variables described above. Robust Huber–White standard errors were used to account for state-level clustering.

Results

Table 1 presents univariate statistics for individual-level variables included in analyses. Across all 50 states, 65.1% of state legislators have a Twitter account, 55.7% of legislators made at least one post using Twitter in 2014 or 2015, and 42.1% posted 500 tweets or more. This indicates that among state legislators, Twitter adoption and use is still a variable, not a constant, and that the adoption of Twitter leads to varying degrees of Twitter use.

Table 1.

Univariate Statistics for Legislator Variables.

Variable	Mean	SD	Min-Max
Account? (1 = yes)	0.651	—	0-1
Active 2014-2015? (1 = yes)	0.557	—	0-1
500 Tweets or more? (1 = yes)	0.421	—	0-1
Republican? (1 = yes)	0.562	—	0-1
Majority party? (1 = yes)	0.648	—	0-1
Upper chamber? (1 = yes)	0.2624	—	0-1
Legislative leader? (1 = yes)	0.1319	—	0-1
Legislative veteran? (1 = yes)	0.783	—	0-1
% of district with Bachelor’s or more	28.867	13.618	3.439-84.393
Median income of district ($thousands)	27.989	7.868	5.511-78.394
Median age of district (years)	38.700	5.176	19.500-63.700

Table 2 presents univariate statistics for the 50 states. The percentage of legislators in each state who have adopted Twitter varies widely, from a minimum of 26.2% in North Dakota to a maximum of 93.3% in California. Nearly half of all states have experienced either a split legislature (with one party controlling each chamber) or a shift in majority control from one party to another between 2010 and 2015.

Table 2.

Univariate Statistics for State Variables.

Variable	Mean	SD	Min-Max
Percentage with account	65.933	16.619	26.240-93.330
Percentage active 2014-2015	52.284	17.925	22.700-90.000
Party control change or split 2010-2015? (1 = yes)	0.480	—	0-1
Percentage of households with Internet access	75.203	17.925	59.879-82.746
Population per legislator (thousands)	42,982	49,546	3116-317224
State Squire Index	0.183	0.116	0.027-0.626

Bivariate associations with Twitter adoption, 2014-2015 use, and high posting activity were separately calculated for each of nine independent variables for each of the 50 states (minus chamber and party variables for Nebraska). The number of associations totals 1,341, too many to summarize in tabular form. Figures 1, 2, and 3 therefore graph the pattern in associations for each combination of independent and dependent variable, counting the number of positive associations and negative associations and categorizing those counts according their degree of statistical significance.

Figure 1.

Bivariate associations with dependent variable “Twitter Account?” (1 = yes) in 50 state legislatures (x-axis: # positive/negative associations in 50 state-level studies).

Figure 2.

Bivariate associations with dependent variable “Active on Twitter 2014-2015?” (1 = yes) in 50 state legislatures (x-axis: # positive/negative associations in 50 state-level studies).

Figure 3.

Bivariate associations with dependent variable “≥500 Tweets by users active in 2014-2015?” (1 = yes) in 50 state legislatures (x-axis: # positive/negative associations in 50 state-level studies).

These figures reveal that different studies of single legislatures would frequently come to starkly different conclusions. Only district education exhibits a statistically significant association with Twitter adoption in at least half of states. All other variables, even adopting the looser p < .10 threshold for significance, appear to be statistically insignificant predictors more often than not.

Low variability may explain some of the insignificant findings. In California, Colorado, Florida, Nevada, Ohio, and Texas, the Twitter adoption rate sits above 85%, not leaving much variation to explain. While some legislatures feature a great many party leaders (in New Jersey, 44.2% of all legislators are listed in leadership), the examples of Louisiana (3.5% in leadership), Alabama (2.1% in leadership), and Mississippi (1.7% in leadership) are more typical. When there are few leaders, the leadership variable shows little variation. Finally, some state legislatures have a small number of legislators, decreasing statistical power.

Despite these limitations, for every variable other than those related to party, there is at least a preponderance of either positive or negative association with Twitter adoption and activity across the 50 United States in Figures 1 and 2. The party variables are an exception, with a number of statistically significant associations appearing in both positive and negative directions for being a Republican and for being in the minority party of a legislature. What being a Republican or being in a minority party means for Twitter adoption and activity is significant in different ways in different states. This pattern is the opposite of what general theories about parties predict.

A final result worth noting is a strong correspondence between Figure 1 and Figure 2 in patterns of significance and direction of effect—with the notable exception of legislative veteran status. Veteran status is an inconsistent predictor of Twitter adoption, tending toward statistical insignificance. By contrast, being a legislative veteran tends to have a more consistent negative association with Twitter use in 2014-2015 that is also slightly more likely to attain at least mild statistical significance. The distinction may be an early sign of Twitter’s waning status as an innovation. A number of legislative cycles have passed since Twitter was adopted as an innovation by political challengers. Having reached office and having attained reelection since, a number of those old-time challengers have become secure veterans and may see no need to actively use Twitter—or delete their account. The debris of old accounts may muddy the relationship between legislative experience and Twitter adoption.

The results of Figure 3 are notably different from those in Figures 1 and 2. Statistically insignificant associations of independent variables with large tweeting volume are far more common, and the direction of effect varies more often as well. The reduction of cases for each state legislature serves a meaningful purpose—to only study variability within the set of active Twitter users—but also reduces statistical power. Even the most commonly significant association (that of district education) is only significant in less than half of states. Trying to explain the emergence of Twitter “super-users” by looking at just one state legislature is not likely to lead to a reliable result.

If Figures 1, 2, and 3 emphasize the variety of patterns in individual states, multivariate models that combine American state legislators into a single-pool emphasize the strength of general trends. The pooling of all legislators (Tables 3 and 4) or all legislators active on Twitter (Table 5) into a single 50-state database adds statistical power and also allows the introduction of state-level variables. Model 1 in each table features variables studied in prior single-state research, while Model 2 adds previously untested state-level variables. The addition of these state-level variables adds considerable explanatory power to models of both Twitter adoption and Twitter activity, as measured by the pseudo R² statistic.

Table 3.

Logistic Regression Models for all Legislators (Dependent Variable = Twitter Adoption).

Independent variable	Model 1	Model 2
Individual level
Republican	0.125	0.139
Majority party	−0.076	−0.079
Woman	0.329***	0.386***
Upper chamber	0.333**	0.340***
Legislative leader	0.444***	0.383**
Legislative veteran	−0.089	−0.169*
District level
District % with Bachelor’s degree or more	0.022***	0.025***
Median income ($thousands)	0.007	0.003
Median age	−0.049**	−0.039***
State level
% Households with Internet access in state		−0.022
Party Switch/Split 2010-2015		0.062
Population per legislator in state (thousands)		0.016***
State Squire Index		1.834*
Intercept	1.710**	1.960
N	7,277	7,277
Pseudo R²	.063	.166

p < .05. **p < .01. ***p < .001.

Table 4.

Logistic Regression Models for all Legislators (Dependent Variable = Active in 2014-2015).

Independent variable	Model 1	Model 2
Individual level
Republican	0.002	0.045
Majority party	−0.121	−0.133
Woman	0.228***	0.273***
Upper chamber	0.331**	0.341***
Legislative leader	0.437***	0.386***
Legislative veteran	−0.332**	−0.421***
District level
District % with Bachelor’s degree or more	0.017***	0.020***
Median income ($thousands)	0.014	0.013
Median age	−0.047**	−0.039***
State level
% Households with Internet access in state		−0.016
Party Switch/Split 2010-2015		0.059
Population per legislator in state (thousands)		0.013***
State Squire Index		2.113**
Intercept	1.309*	1.317
N	7,277	7,277
Pseudo R²	.066	.165

p < .05. **p < .01. ***p < .001.

Table 5.

Logistic Regression Models for Active Twitter Users (Dependent Variable = At Least 500 Tweets Posted).

Independent variable	Model 1	Model 2
Individual level
Republican	−0.104	−0.096
Majority party	−0.141	−0.137
Woman	−0.172*	−0.141*
Upper chamber	0.226*	0.230*
Legislative leader	0.080	0.121
Legislative veteran	0.199*	0.212**
District level
District % with Bachelor’s degree or more	0.014***	0.016***
Median income ($thousands)	0.003	0.005
Median age	−0.026***	−0.024*
State level
% Households with Internet access in state		−0.021*
Party Switch/Split 2010-2015		0.145
Population per legislator in state (thousands)		0.005
State Squire Index		0.850
Intercept	0.1141	1.017
N	4,053	4,053
Pseudo R²	.033	.063

p < .05. **p < .01. ***p < .001.

The lack of a strong effect of the percentage of households with Internet access (or significant negative relationship, in the case of high Twitter activity) may seem counterintuitive, but may be due to the measure’s overly broad measurement at the state level. Even states with low Internet access rates have pockets of high Internet access, and vice versa. When American Community Survey data on household Internet access are made available for state legislative districts, it should replace the state-level variable for a better test.

In contrast, two remaining state-level variables are significantly associated with both Twitter adoption and activity in the predicted direction. The more constituents each legislator is responsible for representing, the more likely that legislator is to adopt Twitter as a strategy for efficient communication at a large scale, then use Twitter in the current legislative term. The more a legislature involves expectations of a significant time commitment but also expectations of staff and resource support, the more likely a legislator in that legislature is to adopt and use Twitter. It is curious that these variables are not significantly associated with the likelihood that a legislator is a high-volume Twitter user, but only with the likelihood that the legislator uses Twitter at all. No current theory of political Twitter adoption is able to explain this distinction.

Turning attention to the legislator/district level, a series of variables predicted by prior studies to have an effect on Twitter adoption prove significant here. Districts with young and educated constituents are disproportionately represented by legislators who have an account, who tweet in the current legislative session, and who tweet a great deal. Median income is not significant as an independent variable, but this may be due in part to the variable’s strong correlation (+.72) with district education, leading to moderate collinearity issues (variance inflation factor 2.6 in Twitter activity models). When pooled into a large enough group to exhibit variation, legislative leaders reveal a tendency not to turn away from Twitter, but to disproportionately adopt and engage with it, as do members of the upper chambers of state legislatures. These deviations from the innovation-based predictions of political science may indicate that in 2015, Twitter was transitioning from a novelty to a more standard option for state legislators. And yet (as with state-level variables) leaders are neither more nor less likely to post a large volume of tweets, a distinction that is unexplained by current theory of political twitter use.

Results regarding legislative veterans remain a noticeable deviation from the overall trend of similarity between predictors of Twitter adoption and predictors of Twitter activity. The weaker, less consistently significant association of legislative newness with adoption paired with the stronger, more consistently significant association of newness with activity suggests again the possibility that old, disused Twitter accounts from when veterans were newcomers may be obscuring the adoption effect. That suggestion, however, is complicated by the finding that legislative veterans are more likely among active Twitter users to be high-volume Twitter users.

In combination, veterans are more likely to have an account, more likely not to use it currently, but if they use it currently more likely to post a great deal. That pattern does not fit neatly into any theory of political Twitter use.

Continuing the trend, an inconsistent gender effect bears notice here. Even controlling for a variety of other individual, district, and state-level variables, and despite the movement of men to Twitter in general in the United States, women in state legislatures remain more likely to adopt and use Twitter than men. However, women legislators who use Twitter are less likely to post in large volumes than their male colleagues. How can the use of Twitter by women state legislators reflect an alternative strategy for gaining voice in a male-dominated environment if those men tend to post more than women do? In interpreting these results, it may be helpful to recall the purposes for which Twitter is created (to follow and be followed) and used as a speech platform (to proclaim statements of position). If these activities are gendered, is it possible that the divergent strengths of men and women legislators on Twitter are also gendered?

Finally, the lack of any significant effect of party switching and splitting on the “innovation” of Twitter adoption in a legislature is consistent with Mayhew’s skeptical position toward Lowi and other electoral alignment theorists. These results provide no evidence that partisan splits or turnover lead to the innovative use of social media. Relatedly, party identification itself remains an inconsistent predictor. As Twitter use leans left in some state houses but leans right in others, the overall effect of party variables across all 50 states for all three dependent variables amounts to a net nil nationally. Considerable unexplained variation in the adoption and use of this innovation nevertheless remains. It may therefore be fruitful in future work to move from general explanations toward contingent ones.

Discussion and Conclusion

The era of variation in political Twitter use is not over. At the state legislative level in the United States, where bills are passed quietly but impactfully into law, Twitter use still varies, being nearly universal in some states and a rarity in others. State-level characteristics are joined by individual- and district-level characteristics in predicting which individual legislators will adopt and use the platform. Constituents per legislator, the youth and educational attainment of a district, legislative professionalism, being a woman, sitting in the upper chamber, leadership, and legislative inexperience are significantly and positively associated with Twitter adoption and current Twitter use. Many other observed associations are fickle, on the other hand, and inconsistent with theoretical explanations offered to date.

Locally significant but broadly inconsistent effects in the 50-state replication discussed here suggest that general explanations for Twitter adoption in American legislatures are insufficient. Low pseudo R² scores for all logistic regression models further reinforce this suggestion; a great deal of variation in Twitter use by state legislatures is not explained by the variables suggested in prior work. Could it be that in the adoption and use of an innovative form like Twitter, general tendencies give way to contingent ones? For different forms of possible contingencies, different approaches to research are called for.

One variety of contingency to consider in future research is social network contingency (Knoke, 1993), in which innovations diffuse along lines of social contact. The experience of two very different legislators and legislatures may be instructive. Democratic Pennsylvania State Representative Mike Schlossberg, a professional social media consultant who has acted as a strong force urging the adoption of social media among his colleagues (Schlossberg, 2016). In Pennsylvania, Democrats make up 41.1% of the legislature but 53.0% of Twitter adopters. On the other side of the country, Republican Oregon State Representative Bill Post guides his colleagues toward Twitter (Marmaduke, 2015). In Oregon, Republicans make up 41.1% of the legislature but 46.2% of Twitter adopters. These substantively different strengths for political parties on opposite coasts feature an underlying commonality of active, social boosterism. To study the spread of Twitter use through social networks, a longitudinal data set tracking the timing of adoption according to changes in patterns of social contact between legislators would be most appropriate.

A second variety of contingency to consider is compositional contingency, in which the relative size of groups affects the expression of identity and behavioral choices of individuals. The drive for legislative minorities to adopt innovation as a strategic reaction to their outsider status is a common theme of this article, but minority status has been considered as dichotomous. Structural differentiation according to group size along multiple dimensions affects the way in which strategic interests form and are articulated (McVeigh, 1995). Recent work on legislative minorities (Childs & Krook, 2008) suggests it is important not only to consider relative group size but also how “critical actors” react to minority status and affect the meaning of that status in turn.

Institutional contingency may provide a third direction for future theoretical development. Unique institutional traditions, schemas, formal procedures, cultural imperatives, and models of action may guide different state legislatures down different paths that constrain or enable innovation in historically contingent ways (Clemens & Cook, 1999), so that what may seem to be structurally equivalent conditions in two legislative bodies may be experienced, confronted, and managed in very different ways. To study social media adoption institutionally is to rely on deep, rich, historically sensitive accounts and to acknowledge that without attention to this context, the ability of a general model to explain variation in social media outcomes may remain limited.

Finally, it may be useful to adopt a similar approach to replication regarding social media use in other structurally similar domains. Do social media habits of registered lobbyists in sets of states, chambers of commerce in sets of regions, corporate boards in sets of industries, and faculties in sets of schools match findings from single case studies? The use of the replication approach demonstrated here may help clarify sporadic, possibly unreliable accounts in these domains as well.

Footnotes

Acknowledgements

I would like to express my thanks to three anonymous referees and the participants of the 2016 International Conference on Social Media & Society for their helpful comments on earlier versions of this article.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Author Biography

James M. Cook (PhD, University of Arizona) is an Assistant Professor of Social Science at the University of Maine at Augusta. His research interests include social interaction in legislatures, the social network analysis of social media, and gender performance in politics.

References

Chi

Yang

(2010). Twitter adoption in Congress. Review of Network Economics, 10(1), 1-46.

Childs

Krook

M. L.

(2008). Critical mass theory and women’s political representation. Political Studies, 56, 725-736.

Clemens

E. S.

Cook

J. M.

(1999). Politics and institutionalism: Explaining durability and change. Annual Review of Sociology, 25, 441-466.

Evans

H. K.

Cordova

Sipole

(2014). Twitter style: An analysis of how House candidates used Twitter in their 2012 campaigns. PS: Political Science and Politics, 47, 454-462.

Gainous

Wagner

K. M.

(2014). Tweeting to power: The social media revolution in American politics. New York, NY: Oxford University Press.

Glassman

Straus

J. R.

Shogan

C. J.

(2009). Social networking and constituent communication: Member use of Twitter during a two-week period in the 111th Congress (Congressional Research Service: Report R40823). Retrieved from https://www.fas.org/sgp/crs/misc/R40823.pdf

Glassman

Straus

J. R.

Shogan

C. J.

(2013). Social networking and constituent communication: Member use of Twitter during a two-week period in the 112th Congress (Congressional Research Service: Report R43018). Retrieved from https://www.fas.org/sgp/crs/misc/R43018.pdf

Golbeck

Grimes

J. M.

Rogers

(2010). Twitter use by the US Congress. Journal of the Association for Information Science and Technology, 61, 1612-1621.

Greenwood

Perrin

Duggan

(2016). Social media update 2016. Retrieved from http://www.pewinternet.org/2016/11/11/social-media-update-2016/

10.

Gulati

G. J.

Williams

C. B.

(2010). Communicating with constituents in 140 characters or less: Twitter and the diffusion of technology innovation in the United States Congress (Social Science Research Network: SSRN 1628247). Retrieved from http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1628247

11.

Knoke

(1993). Networks of elite structure and decision making. Sociological Methods & Research, 22, 23-45.

12.

Lassen

D. S.

Brown

A. R.

(2010). Twitter: The electoral connection? Social Science Computer Review, 29, 419-436.

13.

Lowi

(1963). Toward functionalism in political science: The case of innovation in party systems. American Political Science Review, 57, 570-583.

14.

Marmaduke

(2015, July 16). Why a few Oregon legislators swear by Twitter—and the rest don’t use it. The Oregonian. Retrieved from http://www.oregonlive.com/politics/index.ssf/2015/07/why_a_few_oregon_legislators_s.html

15.

Mayhew

D. R.

(2000). Electoral alignments. Annual Review of Political Science, 3, 449-474.

16.

Mayhew

D. R.

(2005). Divided we govern: Party control, lawmaking, and investigations, 1946-2002 (2nd ed.). New Haven, CT: Yale University Press.

17.

McVeigh

(1995). Social structure, political institutions, and mobilization potential. Social Forces, 74, 461-485.

18.

Moran

C. T.

(2013). Understanding Wisconsin legislators’ use of social media (University of Wisconsin-Milwaukee Theses and Dissertations Paper 299). Retrieved from http://dc.uwm.edu/cgi/viewcontent.cgi?article=1304&context=etd

19.

Peterson

R. D.

(2012). To tweet or not to tweet: Exploring the determinants of early adoption of Twitter by House members in the 111th Congress. Social Science Journal, 49, 430-438.

20.

Reingold

Smith

(2014). Legislative leadership and intersections of gender, race and ethnicity in the American states (Social Science Research Network: SSRN 2487604). Retrieved from http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2487604

21.

Sala

J. F. A.

Jones

M. P.

(2012). The use of electronic technology and legislative representation in the Mexican and U.S. states: Nuevo León and Texas (Puentes Consortium Project). Retrieved from https://www.researchgate.net/publication/303721951_The_Use_of_Electronic_Technology_and_Legislative_Representation_in_the_Mexican_and_US_States_Nuevo_Leon_and_Texas

22.

Schlossberg

(2016). Mike Schlossberg’s Biography. Retrieved from http://www.pahouse.com/Schlossberg/About/Biography

23.

Shpayer

(2016). GovSM. Retrieved from http://www.govsm.com/

24.

Squire

(2007). Measuring state legislative professionalism: The Squire Index revisited. State Politics & Policy Quarterly, 7, 211-227.

25.

Stahl

(2016, November 13). 60 Minutes interview with Donald Trump. 60 Minutes. Retrieved from http://www.cbsnews.com/news/60-minutes-donald-trump-family-melania-ivanka-lesley-stahl/

26.

Sunlight Foundation. (2015). Open states: Bulk downloads. Retrieved from http://openstates.org/downloads/

27.

U.S. Census Bureau. (2009-2014). 2014 American Community Survey 5-year estimates. Retrieved from http://factfinder.census.gov

28.

Warnock

(2015) State partisan composition. National Conference of State Legislatures. Retrieved from http://www.ncsl.org/research/about-state-legislatures/partisan-composition.aspx

29.

Williams

C. B.

Gulati

G. J.

DeLeo

R. A.

(2013). The dissemination of social media to campaigns for state legislature: The 2012 New England case. 2013 Annual Meeting of the American Political Science Association. Retrieved from http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2301245