Staffers as network bridges

Staffers can coordinate legislative action by serving as network bridges, connections between otherwise disconnected nodes in a network, and between legislators (Burgat Reference Burgat2020; Montgomery and Nyhan Reference Montgomery and Nyhan2017; Ringe et al. Reference Ringe, Victor and Gross2013). To serve as network bridges, I hypothesize staffers must (a) have specialized knowledge gained through intensive work, (b) have a large network by working extensively in different roles, and (c) be trusted by their principals.

Staffers have a myriad of different roles because legislators delegate responsibilities to staffers due to time constraints (DeGregorio Reference DeGregorio1995). Legislatures could not operate without the ‘invisible’ labor of staffers (Fox and Hammond Reference Fox and Hammond1977). Staffers control access to the legislator by interest groups (Furnas et al. Reference Furnas, LaPira, Hertel-Fernandez, Drutman and Kosar2023; Jenkins, Landgrave, and Martinez Reference Jenkins, Landgrave and Martinez2020; Kalla and Broockman Reference Kalla and Broockman2016) and by constituents (Hertel-Fernandez et al. Reference Hertel-Fernandez, Mildenberger and Stokes2018). Staffers give advice on policymaking (Gray and Lowery Reference Gray and Lowery2000; Pertschuk Reference Pertschuk2017; Weissert and Weissert Reference Weissert and Weissert2000) and play a major role in constituency service (Frantzich Reference Frantzich1985; Landgrave and Weller Reference Landgrave and Weller2020). These extensive roles allow staffers to build large networks with other legislative actors. Additionally, in the process of carrying out intensive work duties staffers develop specialized knowledge about legislative activity that increases their value to legislators. Specialized knowledge can come in the form of public policy expertise and/or knowledge about how the legislature operates. Working extensively (working in many areas) and intensively (developing specialized knowledge) enables staffers to build an extensive network across the legislature and develop specialized knowledge.

Moreover, staffers must be trusted by legislators to be effective network bridges. Untrustworthy staffers with specialized knowledge would be an issue for legislators because they could influence legislators to act against their own best interests. Rational legislators would not allow staffers to develop specialized knowledge if they could not trust them. A rational legislator will only trust a staffer and delegate duties to them if they can minimize agency loss. One way to minimize agency loss is through monitoring and screening (Kiewiet and McCubbins Reference Kiewiet and McCubbins1991). Legislators are in regular contact with staffers (Fenno Reference Fenno1978) and this gives legislators the opportunity to identify staffers who can be trusted with delegated power (DeGregorio Reference DeGregorio1995; Hagedorn Reference Hagedorn2015).

At the national level, there is strong evidence of the influence of staffer networks on legislative behavior. Montgomery and Nyhan (Reference Montgomery and Nyhan2017) find that Members of Congress (MCs) who have exchanged senior and policy staffers are more similar in their voting patterns than would otherwise be predicted. Montgomery and Nyhan (Reference Montgomery and Nyhan2017) hypothesize that Congressional staffers can influence legislative behavior because their specialized knowledge affords them autonomy which can lead to agency loss. MCs attempt to minimize agency loss by monitoring, retaining, and promoting trustworthy staffers. MCs would prefer to only delegate duties to highly trustworthy staffers but resource restraints force MCs to delegate autonomy to less trustworthy staffers who may use their autonomy to pursue their own goals.

Burgat (Reference Burgat2020) finds that better-connected legislative staffers are more likely to be assigned to important policy assignments because legislators believe they will be better able to leverage their networks to achieve success. McCrain (Reference McCrain2018) tests staffers’ influence in networks by looking at their success as lobbyists and finds that former staffers with more congressional ties are more successful as lobbyists. The findings of Burgat (Reference Burgat2020) and McCrain (Reference McCrain2018) are important because they help us understand why staffers are motivated to serve as network bridges. By increasing their connectedness in the legislative network, staffers not only improve the efficiency of their MCs (Fowler Reference Fowler2006; Battaglini, Sciabolazza, and Patacchini Reference Battaglini, Sciabolazza and Patacchini2019) but also improve their own future wages as lobbyists.

Much of the extant work on staffer networks has relied on data from the United States House of Representatives (Fox and Hammond Reference Fox and Hammond1977) and Senate (Pertschuk Reference Pertschuk2017), but there is evidence that staffers serve as network bridges in the comparative context as well. Ringe et al. (Reference Ringe, Victor and Gross2013) have noted the influence of staffers in the European Parliament. While staffer networks influence legislative behavior in the US and abroad, it is unclear if the influence of personal staffers extends to state legislatures where institutional design may limit the influence of staffer networks.

Staffer utilization in state legislatures

State legislative staffers do not have the same roles, and are less professionalized, than their Congressional peers, and this may hinder their potential as network bridges. It cannot be taken for granted that staffers in all legislatures are used intensively and extensively. Nor can it be taken for granted that state legislative staffers are trusted by state legislators to the same extent Congressional staffers are trusted by MCs. American legislatures differ on many institutional design elements (Squire and Hamm Reference Squire and Hamm2005; Squire Reference Squire2024), such as whether their legislators are term-limited (Kousser Reference Kousser2005) and or whether their members are elected from single or multimember districts (Kirkland Reference Kirkland2012). I argue that the influence of legislative staffers is conditional on the variation in human resource management and staffer utilization across American legislatures.

Staffers have historically been treated in the literature as agents with minimal agency loss, that is, staffers are assumed to be acting on behalf of their MCs (DeGregorio Reference DeGregorio1988). This assumption is so pervasive that many legislative studies purported to be studying legislators use data on staffers as proxies for legislators. However, the principal–agent model cannot be taken for granted as accurately representing the relationship between staffers and legislators in state legislatures. State legislative staffers are employed by a centralized human resource department which severely constrains the relationship between them and legislators.

This constraint is perhaps most blatant when looking at how staffers are assigned in state legislatures. In Congress, MCs have a high degree of power to select and retain their staffers. In comparison, in some state legislatures, staffers are given assignments by centralized HR departments. In Arizona and New Mexico, staffer assignments change every two-year session. The new assignment rate, the percentage of staffers that are assigned new legislators each session, is about 100 and 85 percent, respectively, in Arizona and New Mexico; see Table 1. In Indiana, staffer assignments are changed every year with a new assignment rate of 74 percent. The high new assignment rate means that staffers have divided loyalties between their current legislator(s) and the human resource department that controls their long-term assignment. State legislative staffers serve at the pleasure of their current legislators, but they also serve as the pleasure of the human resources department. If staffers desire long-term employment, they must carefully balance their loyalty between their two principals.

Table 1.

Institutional features of the US, Arizona, Indiana, and New Mexico house of representatives

Note: Staff Turnover for the US House is the average for 2001–2018. Arizona and Indiana Houses are for 2015–2020. New Mexico is for 2017–2020.

Further complicating the principal–agent model is that personal staffers in state legislatures like Arizona, Indiana, and New Mexico have multiple legislator principals. Their position as legislators’ trusted confidants creates a problem as it means that they are privy to multiple legislators’ confidential information. Access to this information could manifest itself as an opportunity for them to serve as network bridges but it could also lead to them betraying the trust of one legislator to aid another legislator. It is vital for legislators to uphold strong norms of conduct for staffers lest order break down. In possible recognition of the problem presented with dual loyalties for shared staffers, legislators have a norm against asking staffers about other legislators. The National Conference of State Legislatures’ (NCSL) suggested code of conduct (NCSL 2019) is evidence of the prevalence of this norm:

Survey evidence also supports the existence of a confidentiality norm. In an original survey of state legislators fielded in Spring 2020 (n = 81), I asked state legislators if they ever asked shared staffers about other legislators’ activities. Among those with shared staff, only 5.40 percent said that they had inquired about other legislators’ actions.

A consequence of the principal–agent model breaking down in state legislatures with shared personal staff is that staffer professionalism is significantly lowered. A rational legislator will not delegate duties to a staffer who is untrustworthy. In turn, a staffer without delegated duties will neither have the ability nor incentive to develop specialized knowledge.

Clerks or political professionals?

One of the core questions in the legislative staffer literature is whether to theorize staffers as “clerks” or “political professionals” (Romzek and Utter Reference Romzek and Utter1997). If staffers are clerks, then they are workers present to extend the capacity of legislators by conducting simple routine tasks and only matter in the sense that other capital inputs matter in determining a legislators’ effectiveness. If staffers are political professionals - individuals with specialized knowledge (Wilensky Reference Wilensky1956) - then staffers matter not only in determining a legislators’ effectiveness but also in so far that they influence decision-making on roll-call votes and other legislative behavior.

There is strong evidence that Congressional staffers are political professionals (Hertel-Fernandez et al. Reference Hertel-Fernandez, Mildenberger and Stokes2018; Montgomery and Nyhan Reference Montgomery and Nyhan2017; Pertschuk Reference Pertschuk2017), but there is less certainty of how to classify state legislative staffers. There are studies on state legislative staffers’ influence (Weissert and Weissert Reference Weissert and Weissert2000), but most extant studies are insufficient to make a generalizable statement because they often focus on a single legislature or a group of legislatures with little variation between them. The study of state legislatures does not require all state legislatures to be analyzed (Nicholson-Crotty and Meier Reference Nicholson-Crotty and Meier2002), but the selected legislatures must be of sufficient range in institutional design if one wishes to make a generalizable claim. A strength of the present study is that its three cases are, respectively, among the most (Arizona) and least (Indiana and New Mexico) professionalized state legislatures with shared personal staff.

One method to measure staffer utilization, whether staffers are clerks or political professionals, is staffer turnover rates (Salisbury and Shepsle Reference Salisbury and Shepsle1981a; Jensen Reference Jensen2011). Turnover rates signify the percentage of staffers that are new in each session. A lower turnover rate signifies that a legislature has a relatively continuous staffer force. Continuity allows staffers to gain specialized knowledge of legislative norms (Matthews Reference Matthews1959; Herrick and Fisher Reference Herrick and Fisher2007) and organization-specific tasks. This specialized knowledge allows staffers to act as political professionals. High turnover on the other hand means that continuity is not guaranteed and discourages staffers from acquiring specialized knowledge through intensive work or from working extensively to become better connected in legislative networks.

Table 1 presents, among other descriptive statistics, staffer turnover rates for the United States, Arizona, Indiana, and New Mexico House of Representatives. Notably, the turnover rate is much higher in legislatures than in the private sector. The United States Bureau of Labor Statistics estimates yearly turnover rates (defined as “quit rates”) in the low single digits for most of the private sector. The relatively high turnover rate among legislatures is because few personal staffers intend for it to be a lifelong profession (Salisbury and Shepsle Reference Salisbury and Shepsle1981a). Many staffers intend to segue into adjacent professions such as lobbying (McCrain Reference McCrain2018; Rosenthal Reference Rosenthal2000) or run for political office themselves (Moncrief, Squire, and Jewell Reference Moncrief, Squire and Jewell2001). Despite the overall high turnover rates among legislatures, there is significant variation across legislatures.

The Arizona House of Representatives has a staffer turnover rate of 38.04 percent. Indiana and New Mexico, respectively, have turnover rates of 39.02 and 75.86 percent each. For comparison, the United States House of Representatives has a personal staff turnover rate of 20 percent according to LegiStorm. These figures suggest that state legislative staffers have significantly less specialized knowledge than their Congressional counterparts because they have had little time to acquire it and or little incentive to develop it.

The differences between national and state legislative staffers are important because shared personal staff may not play as important a role as network bridges in less professionalized legislatures with high turnover rates. High turnover rates limit staffers’ ability to build specialized knowledge through intensive work or connections through extensive work that would allow them to maximize their potential as bridges in the legislative network.

In the United States House of Representatives there are 20 personal staffers per legislator. In the Arizona and New Mexico House of Representatives there are 0.5 staffers per legislator. In Indiana, there are 0.33 staffers per legislator. The low numbers of staffers in state legislatures places a burden on state legislators that can lead to conflict over staffer utilization.

One way to deal with the scarcity of staffers is to create clear rules on who can utilize a staffers’ time in each point of the day. While conducting background research for this manuscript I interviewed personal staffers in Arizona, Indiana, and New Mexico. Many stated that they usually divided their day around lunch time. They’d work for legislator A in the morning and then work for legislator B in the afternoon. In some instances, staffers would physically move between the offices of legislators A and B. While this arrangement discourage antagonism between legislators by setting clear property rights to staffers’ time (Coase Reference Coase1960), it does not facilitate coordination. One can imagine a story where legislator A (B) is in their office in the morning (afternoon) to coincide with their staffer and spends the remainder of their day elsewhere. This would mean that legislators A and B may be less likely to see each other in each day than if they did not share a staffer.

Research design and discussion

In this article, I test if shared personal staffers networks are associated with legislators’ state legislators’ roll-call voting. I hypothesize that personal staffers may promote congruent roll-call voting by serving as a network bridge between the legislators they share. Network bridges can be brokers of information between groups that otherwise have minimal to no contact with one another. Network bridges serve to coordinate behavior in this way. To serve as network bridges staffers must (a) develop specialized knowledge through intensive work, (b) have an extensive network through extensive work, and (c) be trusted by legislators.

If shared personal staffers are network bridges, then legislators with shared staff should be positively associated with roll-call voting congruency. That is, Legislators who share staffers should be more likely to vote the same way than legislators who do not share staffers. Conversely if staffers are network gaps, legislator with shared staffers should be less likely to vote the same way.

For empirics I rely on an original dataset of state legislative staffers serving in the Arizona (2015–2018), Indiana (2015–2018), and New Mexico (2017–2018) House of Representatives. The selected years are based on the availability of staffer network data. This dataset was created by web scrapping state legislative websites and from FOIA requests. Commercial providers exist for congressional staffer data, but historical data on state legislative staffers are rare. I have selected these three states because they represent a range of the most (Arizona) and least (Indiana and New Mexico) professionalized state legislatures as measured by both Squire (Reference Squire2017)’s professionalism index and staffer turnover rates. If I find or do not find that staffer networks are associated with roll-call vote congruency in all three cases, then I can be more confident in generalizing my results.

I append the original dataset of state legislative staffers with roll-call data from Open States, a nonprofit that has collected roll-call data for state legislatures since approximately 2010. This combined data allows me to test the influence of shared staff on congruent voting.

Vote congruency is a binary outcome that indicates when members of a dyad vote the same way, whether in favor, against, or abstaining (=1) on a given piece of legislation. Vote congruency as an outcome measure can be traced back to Rice (Reference Rice1925)’s proposed cohesion index for the study of legislative behavior. Congruency measures are commonly used in contemporary legislative network studies (Ringe et al. Reference Ringe, Victor and Gross2013; Parigi and Bergemann Reference Parigi and Bergemann2016; Harmon, Fisman, and Kamenica Reference Harmon, Fisman and Kamenica2019; Wojcik Reference Wojcik2018).

I converted my data into dyads between all possible pairs of legislators; see Table 2 for an example of the data frame. In some cases, missing data prevented the creation of a dyad. Note that dyads (ij) are embedded within legislative bills (l).

Table 2.

Sample data frame

To further help understand the data, summary statistics of the legislator-legislation dyads (Table 3) and a description of how the variables were coded (Table 4) are presented below. Within the dataset, approximately 72-80 percent of the dyads are shared votes, which means that legislators voted congruently. Approximately between 0.8-5.1 percent of dyads involve a shared personal staffer. Note that not all variables are available in all three states of Arizona, Indiana, and New Mexico due to variation in data availability.

Table 3.

Summary statistics

Table 4.

Description of variable coding

One of the great difficulties with analyzing network data is that network formation is endogenous; see Supplementary Appendix 1. Given the absence of a randomized intervention such as a lottery to determine staffer assignment (Rogowski and Sinclair Reference Rogowski and Sinclair2012), I refrain from making a strong causal claim. Conversely, if I do not find that dyad pairs with shared staff are more likely to vote congruently despite endogeneity, I am more confident in the null result. Regardless, data permitting, I account for potential confounders such as whether a dyad pair are from the same multimember district (available in the Arizona sample) or if they are co-ethnics (available in the New Mexico sample).

In Table 5 I present the following linear probability model (LPM): $ \mathrm{Congruent}\;{\mathrm{RollCall}\ \mathrm{Vote}}_{\mathrm{ij}\mathrm{l}}={\mathrm{B}}_0+{\mathrm{B}}_1\;{\mathrm{Shared}\ \mathrm{Staffer}}_{\mathrm{ij}}+{\mathrm{B}}_{\mathrm{h}}\;{\mathrm{H}}_{\mathrm{ij}} $ for the Arizona, Indiana, and New Mexico House of Representatives, respectively. Standard errors are clustered by dyad pair––as it standard in dyadic analysis (Green, Kim, and Yoon Reference Green, Kim and Yoon2001). Probit results are reported in Supplementary Appendix 3––results are substantively the same as their LPM counterparts.

Table 5.

Association of shared staff with congruent Roll-Call voting, LPM

Note: Standard errors are in parentheses and are clustered by dyad pair. State legislature fixed effects accounted for in Column 4. *p < 0.1; **p < 0.05; ***p < 0.01.

In Table 5 column 1, I fail to find a relationship between shared staff and vote congruency in the Arizona House of Representatives (p = 0.672). Similarly, in Table 5 column 2 I fail to find evidence (p = 0.508) that shared staff in the Indiana House of Representatives is associated with vote congruency. In Table 5 column 3, I find that shared staffers are negatively associated with roll-call vote congruency (p <0.001) in the New Mexico House of Representatives. In Table 5 column 4 I pool the data from the three respective legislatures and account for state legislature fixed effects. Pooled results find that shared staffers are associated with decreased congruent voting (p-value <0.001).

It is worth noting that in all observed legislatures, staffers are only shared among partisans. Based on the pooled results from Table 5 column 4, this means that the baseline vote-congruency is approximately 67.1 percentage points for legislator dyads that does not share a political party and does not share a personal staffer. Vote congruency is approximately 83.8 percentage points when a legislator dyad shares a political party but does not share a personal staffer. Voting congruency is 77.2 percentage points when a legislator dyad shares a political party and a personal staffer. I do not observe an instance where a legislator dyad does not share a political party and does share a personal staffer.

The negative results may be evidence that staffers serve as network gaps. As discussed above, there is evidence that state legislative staffers adhere to a norm of not divulging information between their respective principal legislators to discourage incidents of staffers violating legislators’ trust. The negative results are supportive of the existence of this norm.

There is debate in the methodological literature on how to deal with standard errors in dyadic analysis. The major issue is that the standard errors do not meet the independent and identically distributed (i.d.d.) assumption, that is, the voting behavior of a dyad in bill A is likely not independent of the dyad’s behavior in bill B. One method of addressing the problem is to cluster the errors by dyad pair, as I have done in Table 5. An alternative method suggested by Erikson, Pinto, and Rader (Reference Erikson, Pinto and Rader2014) is to use randomization inference. Randomization inference does not require the i.d.d. assumption and instead produces a p-value by generating a distribution of the hypothetical coefficient of interest (Keele, McConnaughy, and White Reference Keele, McConnaughy and White2012). I conduct randomization inference using Heß (Reference Heß2017)‘s package for Stata to compute 1,000 hypothetical distributions of p-values. Results are presented in Table 6. Covariates are suppressed for brevity.

Table 6.

Association of shared staff with Congruent Roll-Call voting, LPM with randomization inference p-values

Note: Covariates are suppressed for brevity.

After using randomization inference p-values, shared staffers in all three observed state legislatures are associated with a decrease in roll-call congruency. Even with the bias in favor of finding a positive association due to staffer assignment being driven in part by homophily, shared staffer networks do not increase coordination between legislators. On the contrary, they appear to decrease coordination as measured by roll-call congruency. Additional robustness checks, including a falsification test, can be found in the Supplementary Appendices.