Financial Interests and Economic Diversity in State Legislatures

Abstract

Objective

I examine how the prevalence of U.S. state legislators economically connected to an industry, and the diversity of economic connections in a legislature, vary in response to the employment share of that industry in the state, the partisan split of the legislature, and professionalization (or its components).

Methods

I use conflict-of-interest filings for 1999–2000, 2000 Census employment data, and related data to examine the prevalence of connections to several sectors using seemingly unrelated regression.

Results

The employment share of an industry in the state is a powerful predictor of connections to that industry among legislators, as is the partisan split of the chamber. Professionalization plays a role, dominated by its pay component, but to a smaller degree. Professionalization is the primary predictor of the diversity of connections.

Conclusion

Legislators’ economic connections are driven in part by underlying conditions such as the economic makeup of the state.

Analyses of legislators’ socioeconomic characteristics (Hyneman, 1940; Wahlke et al., 1962) are among the earliest quantitative works on state legislatures. This early work showed uniformly that attorneys, farmers, merchants, and “professional men” served out of proportion to their population numbers, while working-class legislators were rare. More recent studies (Squire, 1992; Bazar, 1987) echoed this finding, with the major change being the growth in “full-time legislator” as an occupation. However, little work has investigated what affects the economic makeup of American legislatures. Squire (1992) found that professionalization, region, and state diversity predicted the prevalence of some external occupations and occupational diversity. Rosenson (2006) found that ethics laws, compensation, and session length affected the prevalence of attorneys and business owners. Beyond these works, little research has examined what affects the industries legislatures draw from. Further, the existing work has omitted a key variable: the economic characteristics of the state. If we ask why there are more legislators connected to fishing in Alaska, Washington, and Oregon than we find in Kansas and Iowa, part of our answer should be that there are more fishermen on the Pacific coast than in the plains.

In this article, I reconsider the distribution of legislators’ economic backgrounds as a function, in part, of their state's employment patterns. Looking at legislators’ 1999–2000 conflict-of-interest filings in 25 lower chambers, I find that the proportion of a legislature connected to an industry is generally well predicted by employment in that sector and the partisan split of the chamber. Professionalization (Squire, 1992, 2000, 2007) has a less-consistent effect, though its pay component acts more strongly. Legislative economic diversity is affected by a state's employment diversity, the size of its majority party, and legislative professionalization (and, again, its pay component).

Literature

A series of early works offered descriptive accounts of legislators’ occupations. Hyneman (1940) found that farmers and attorneys were the most common legislative occupations, that legislators held many occupations, and that working-class occupations were rare. Zeller (1954) also offered a primarily descriptive account of legislators’ occupations. Zeller found that in 1949, attorneys, farmers, and businessmen remained the most common occupations in the statehouse (Zeller, 1954:70–74). Finally, Wahlke et al. (1962) found that over a third of legislators in the states they examined were attorneys, and that insurance, retail trade, and agriculture were also common occupations. While primarily descriptive, this early literature does include some (implicit) analysis of interstate variation. For example, Hyneman (1940) compared the proportion of a legislature with a particular occupation to the proportion of its parent population but offered no further analysis. Wahlke et al. (1962) noted that the degree to which members of a given occupation have control over their own hours, which assists them in combining their external work with their legislative work, is related to an industry's presence in the statehouse. (1962:110)

A few more recent pieces have examined cross-state variation in economic backgrounds. Squire (1992) found that professionalization, region, and state diversity (Sullivan, 1973) played varying roles in legislatures’ occupational makeup, and that professionalization was associated with lower occupational diversity. Rosenson (2006) found that increases in legislator compensation or requiring legislator-attorneys to reveal their clients reduced the proportion of lawyers, and that increasing session lengths were associated with fewer business owners. Carey, Niemi, and Powell (1998) and Carey et al. (2006) found that term limits had no discernible effect on the occupational composition of state legislatures. Maddox (2004a, 2004b) found that men, Republicans, younger legislators, and more educated legislators were more likely to have some external career, reflecting their opportunity costs of service and the likelihood of their already holding a lucrative career in which to continue. Maddox (2004a) found that higher pay was associated with fewer legislators holding external careers, as simple cost-benefit calculations would indicate. My own findings that higher pay and more Democratic chambers are associated with more legislators without coded connections to any industry support these findings.1 However, Maddox was concerned with whether or not a legislator has an active and ongoing external career, but not with which industry that career is in. I analyze which industries legislators are connected to, even if that connection is not part of an active external career.

Pitkin (1967) offers some reasons to be concerned with how closely legislators’ economic backgrounds resemble the larger economy of the state. As Pitkin notes, many theorists have suggested that legislatures ought to resemble their polities, if not perfectly than within some acceptable bound. A particular concern lies between pure descriptive representation and substantive or “acting-for” representation (Pitkin, 1967:112): if legislators do not sufficiently resemble the people they represent, they may act for themselves and for people that they do resemble instead of acting for constituents that they do not resemble. Wahlke et al.'s (1962) concept of the politico illustrates this concern even with elections holding legislators accountable. A politico acts as a delegate on issues where constituents electorally demand a certain vote. On other less-salient issues, however, the politico votes his or her preferences. In state legislatures, low levels of voter information and, often, low levels of electoral competition should expand the issues where politicos can vote their preferences, relative to the U.S. House. And we should expect that legislators’ preferences are in part derived from their economic attachments. This need not be self-aggrandizement, though of course that can occur.2 Rather, all that is needed is for legislators’ preferences about what constitutes good policy to be affected by their economic attachments.

Further, there is empirical evidence in support of this concern. Dyer (1976) found that attorneys in three legislatures were more likely to vote against no-fault auto insurance. This finding is consistent with both self-aggrandizement (practicing attorneys might lose income if fewer suits are filed) (1976:453) and simple differences in preferences (attorneys may sincerely believe no-fault insurance to be bad policy). Ultimately, there is some evidence that legislators’ economic connections affect legislative behavior. This leads to a concern that a legislature whose economic backgrounds differ “too” strongly from those of the state as a whole may limit responsiveness and policy congruence on less-salient policies, where congruence is already less likely (Lax and Phillips, 2012).

A Simple Model of Legislators’ Financial Connections

The economic connections of state legislators vary quite strongly from state to state. The proportion of legislators connected to agriculture ranges from 2 to 25 percent, those connected to law range from 6 to 29 percent, and those connected to real estate and leasing varies between 11 and 67 percent. What processes might lead to such variation? I offer a very simple three-stage model that stresses the costs and benefits of legislative service. This model assumes that first, people acquire interest in legislative service. Second, they look at the costs and benefits of service and decide whether or not to pursue it. Finally, the electorate accepts some candidates and rejects others.

Stage 1: The Supply of Potential Candidates

In the first stage, Nature randomly selects some people to become sufficiently interested in the legislature to consider, however minimally and momentarily, serving in the legislature. This interest need not be intense or sustained. This basic raw interest in service may actually be correlated with economic connections. However, so long as any correlations that exist are consistent from state to state, they should not have any great effect on the cross-state variation I examine. The first stage creates a pool of potential candidates that is a random sample of the adult population and so should resemble it.3 In states with more farmers, more of the people who become interested in the legislature should be connected to farming, and so on. And ceteris paribus we should expect the characteristics of this pool of potential candidates to affect the characteristics of actual legislators, if imperfectly. This leads to my first hypothesis:

Hypothesis 1: All else equal, the proportion of a legislature connected to a given industry should be positively related to state employment in that industry.

It may seem obvious that the prevalence of a given economic sector in the legislature should be at least loosely related to the prevalence of that sector among the population. While socioeconomic diversity (Squire, 1992) or occupational diversity (Rosenson, 2006) have appeared as predictors of legislators’ external occupations, this study is to the best of my knowledge the first that examines this direct connection.

Stage 2: The Costs and Benefits of Service

In a second stage, the potential candidates consider—again, however briefly and minimally—the costs and benefits of legislative service and decide whether to stand for election. The costs of service are significant. Compensation remains low except in the few most professionalized bodies and even the briefest service takes legislators away from external work for a month or more. For many potential candidates, this investigation is therefore quite easy: if being absent from one's principal employment for a few months per year would result in being fired, and legislative pay would not replace lost wages, one is likely to reject legislative service.

Many have asserted that the costs and benefits of service affect the membership of legislatures, at least at the margins. For example, early studies noted the abundance of lawyers in state legislatures (Wahlke et al., 1962; Jewell and Patterson, 1966; Derge, 1959, 1962). One reason offered for their abundance is that legislative service could be financially advantageous to an attorney, unlike most other occupations. Service could provide income to new attorneys when clients might be thin (Wahlke et al., 1962; Derge, 1959, 1962), and present opportunities to network in the capital (Barber, 1965). Jewell and Patterson (1966) came to a similar conclusion. In a similar vein, Wahlke et al. noted that insurance agents and real estate agents were also common, and attributed this to their greater ability to control their own hours (Wahlke et al., 1962:110). A real estate agent has a greater ability to work a partial schedule or to be away from the office for weeks at a time without being fired than does a machinist or office clerk. More recently, Squire (1992) found that less-professionalized legislatures had more students, homemakers, and retirees—groups that do not (usually) depend on their own work in the market and so do not face the same financial opportunity costs of service. Fiorina (1994, 1999) extended this argument, writing that the classic amateur or citizen legislatures favor lines of work that are flexible enough to be combined with legislative service (and people not working in the market), while professionalized legislatures favor those who are willing to give up an external career for legislative service. While his conclusions regarding divided government remain controversial (see Squire, 1997; Fiorina, 1999) the key factor here is the idea that variation in the costs of legislative service can affect the industrial makeup of legislatures.

The most obvious set of cost-benefit variables are clearly legislative professionalization (Squire, 1992, 2000, 2007) and its components. Squire's professionalization index combines a legislature's salary, session length, and staff support and compares them to the U.S. House. Early proponents of professionalization (i.e., CCSL, 1971; Burns, 1971) argued that higher pay would allow middle-class legislators to serve without unduly harming their financial life, opening the doors of the statehouse to more occupations and so increasing diversity. Squire (1992) found that more professionalized legislatures had fewer homemakers, students, and retirees; fewer business employees; and more full-time legislators. The net effect of these changes was a less occupationally diverse legislature. This leads to a reasonably clear prediction for the 1995–1996 Squire index as a whole:

Hypothesis 2: Higher professionalization should be negatively associated with connections to those sectors privileged by a citizen legislature because of greater control over time. These include agriculture,4 insurance, real estate, and law.

However, as with any career decision, potential candidates may be especially responsive to compensation. By breaking the index into its components, it is possible to derive a fuller set of hypotheses. Pay and session length should operate as predicted for the full index. Increased pay reduces the costs of service, reducing the premium on flexible external employment, while longer sessions imply less time to work in that external employment. Both of these factors should, therefore, act to reduce the prevalence of connections to those relatively flexible lines of work.

Hypothesis 3: Higher compensation should be negatively associated with connections to those flexible sectors privileged by a citizen legislature.

Hypothesis 4: Longer sessions should be negatively associated with connections to those flexible sectors privileged by a citizen legislature.

Staff support increases the informational independence and information processing capacity of the legislature instead of affecting the financial burden of service. It enhances the explicitly political aspects of legislative service in a way that pay does not. This different mechanism of action makes generating a theoretical expectation difficult. By increasing the informational capacity of the chamber, higher staff support might act to increase the aggregate level of interest in policy making in the chamber. However, if having a more intense interest in policy making is correlated with the sectors a legislator is connected to, these correlations are far from obvious. Therefore, while I have left staff support in my models where Squire's professionalization index is broken up, I treat it as a control. Models run without staff support are similar to those reported here.

Stage 3: Selection by the Electorate

Finally, the eventual candidates present themselves to the electorate. Most voters are unlikely to be aware of their legislator's or challenger's economic connections as they know little about their legislatures overall. For example, in 2003, Kurtz, Rosenthal, and Zukin found that only 33 percent of adults over 26 knew even which party controlled their state legislature, so the proportion who know specific information about specific legislators should be low as well. It seems unlikely, then, that voters commonly filter candidates on their economic connections. However, voters clearly do filter candidates on the basis of party. And, as Fiorina (1994) notes, party is correlated with connections to the economic world. Fiorina argued that some economic backgrounds trend Republican (farmers, business owners, independent professionals) and others trend Democratic (union officials, teachers, public-sector, and nonprofit workers). This implies that states with more Democratic legislators are, as a side effect, selecting for economic connections that trend Democratic and away from industries that trend Republican. In an ideal world, this partisan screen would generate a hypothesis that legislatures that were more Democratic should be expected to have, ceteris paribus, more legislators connected to some specific industries. However, the diversity of the U.S. economy makes determining the partisan mix of a given industry with any real precision difficult because the resulting industry samples are small, even in large surveys. Lacking the necessary data to generate a useful hypothesis here, I treat the chamber's percent Democratic as a control variable. If two states have similar employment patterns but one elects more Democrats for whatever reason, that state should have more legislators connected to industries where Democrats are more common, whatever those might be, so accurately assessing the effect of population employment shares or professionalization requires controlling for the partisan split of the legislature.

To see the model in practice, consider agriculture-connected legislators in Kansas. Agricultural employment in Kansas is high at 3.3 percent, suggesting that in Stage 1 there are more farmers in Kansas to acquire an initial interest in the legislature than in many states. Pay in Kansas is low at $4,048 and the session is short at 64 days, so people who are more able to temporarily step away from their external work should be less likely to reject legislative service as too great a burden—in Stage 2, fewer agriculture-connected potential candidates filter themselves out. Finally, if we assume for argument's sake that farmers trend Republican, the Republican majority indicates that fewer agriculture-connected legislators are filtered out by the electorate in Stage 3. Together, these factors should combine to create a legislature with strong connections to farming, and indeed almost 25 percent of the legislature is so connected.

Data

Dependent Variable: Legislators’ Economic Connections

The existing literature has used self-stated occupation to measure legislators’ connections to different industries. Squire (1992), Rosenson (2006), and early work operationalized occupation through legislative directories or related documents, while Carey, Niemi, and Powell (1998) included an occupation question in their survey. While having high face validity, occupation does face some inferential problems. Occupation is self-reported and may not accurately reflect the legislator's economic attachments. Even if few voters would be aware of their legislator's occupation as stated in the directory, risk-averse legislators might nonetheless avoid stating occupations that they view as electorally unhelpful. Squire (1992) noted that while observers considered the overwhelming majority of Michigan legislators to be full-time legislators, not one legislator reported his or her occupation as “legislator.” A second problem with occupation is that some stated occupations, such as “retiree,” do not reveal very much about who the legislator resembles. A retired steelworker, a retired K-12 teacher, and a retired banker resemble different subsets of the population. Further, someone who is now a full-time legislator who had previously worked in medicine might differ from one who had worked in finance. A third problem with occupations as a guide to interests and preferences is that occupations are (normally) exclusive. A legislator who is an attorney is not a farmer or a teacher. This coding obscures connections across more than one industry. Legislators may own multiple businesses across different industries, or might receive employment income in one industry and retirement income from another, or might supplement their employment income with business income from another sector.

Instead of occupational data, I measure legislators’ economic backgrounds using their conflict-of-interest filings with the state. Maddox (2004a, 2004b) used an earlier version of the same source data I use in his studies. These economic connections have several advantages. First, they are legally required, sometimes with punishment for misstatement or omission, so we can be confident that the economic connections they report are not merely the connections they wish to be seen to have. Second, these filings have the potential (not always realized) to capture the economic attachments of retirees and full-time legislators. Finally, the conflict-of-interest filings allow legislators to be connected to both multiple sectors. Forty-three percent of legislators in my sample of states have at least two connections, and 5.8 percent have four or more. The financial connection data I use were originally collected by the Center for Public Integrity (2006). The Center collected all of the state legislative conflict-of-interest filings in 1999 and coded each reported interest into 11 categories. I examine the following economic sectors, generally following the Center's coding scheme: agriculture, health care, education, law firms, transportation, mining and oil extraction, communications and electronics, construction, hospitality, finance and insurance, real estate and leasing, general business, and those legislators who report no coded financial connections to any of these industries.5 I also separately coded legislators who had no financial connections coded into the data. A legislator with no coded interests may have meaningful uncoded interests, such as investments.

I also generated an economic-connection diversity index. Because economic connections are not exclusive, the obvious Herfindahl index would be inappropriate. I use the same method as Sullivan (1973) in generating his diversity index. I recorded the proportion of legislators connected to each industry as one category, π₁, and the proportion not connected to it as another, π₂. The index is then , where K is the total number of categories and I the total number of industries.6 The raw variable takes a narrow range, so for expository convenience I rescaled it to range from 0 to 1 in the data. Higher scores indicate greater diversity.

The financial interest data as I use them are binary. Obviously it would be preferable to know that one legislator's connection to agriculture was larger than another's, but this information is almost always unavailable in the original filings themselves or is too vague to be useful. The data include income from employment, retirement, farm or business activity, officerships or directorships, personal business interests, sales and commissions, and property rental.7 By limiting the coding to these income categories and excluding investments,8 the resulting connections should represent meaningful attachments to industries. The data include the legislators’ own financial connections and those of his or her spouse and immediate dependents where such reporting was required. The primary reason for this coding was to avoid losing the substantial minority of filings that do not indicate whether the connection was the filer's, spouse's, or a dependent's.

Filing requirements vary between states. The Center examined legal requirements for filing, the minimum size of interests that must be reported, potential punishments for failing to file, and other factors and combined them into an “openness” rating for each state. Twenty-six lower chambers received “passing” ratings, and I consider all of these states except South Carolina, which had a very high proportion of legislators with no coded connections. The resulting nonrandom sample of states is skewed toward more professionalized states (p = 0.08, two tailed, in a logit model of sample membership using the 1995/1996 Squire index), more populous states (p = 0.11), and more diverse states (p = 0.03 using a 2000 version of the Sullivan index), but there were no statistically discernible differences between sampled and unsampled states in majority size, Democratic majority, term limitation, or leader institutional power using the Clucas (2001) index. Among the sample of states, the primary ways reporting requirements varied were whether or not they required the disclosure of spouse and dependent income, and whether lawyer-legislators were required to reveal their clients.

Table 1 presents the percentage of legislators in each chamber connected to each of the various sectors, the percentage with no coded connections, the economic diversity index, and the state's reporting requirements as coded by the Center for Public Integrity.

Table 1. Distribution of Legislators’ Economic Connections, 1999–2000

State	AK	AL	AR	AZ	CA	CO	CT	DE	FL	HI	KS	KY	MA
Agriculture	7.5	11.4	13.5	15.0	6.3	10.8	7.3	7.3	10.1	2.0	24.8	22.0	3.1
Health care	12.5	8.6	12.5	11.7	12.5	9.2	10.6	0.0	6.7	17.6	12.8	15.0	13.1
Education	12.5	33.3	21.9	23.3	12.5	29.2	19.2	19.5	11.8	31.4	18.4	29.0	21.3
Law	10.0	6.7	6.3	13.3	13.8	16.9	19.9	9.8	15.1	19.6	8.0	20.0	25.6
Finance and insurance	17.5	20.0	17.7	28.3	11.3	20.0	16.6	14.6	15.1	23.5	20.0	24.0	11.9
Real estate	67.5	43.8	16.7	41.7	33.8	27.7	31.1	24.4	26.9	35.3	26.4	32.0	17.5
Transportation	17.5	4.8	8.3	6.7	1.3	6.2	4.6	0.0	4.2	5.9	6.4	8.0	0.6
Mining	10.0	1.0	3.1	5.0	1.3	1.5	0.0	0.0	0.8	2.0	5.6	2.0	1.9
Communications and electronics	7.5	5.7	1.0	3.3	5.0	9.2	6.6	2.4	2.5	3.9	4.8	5.0	4.4
Construction	17.5	7.6	4.2	8.3	6.3	6.2	4.0	2.4	2.5	2.0	5.6	10.0	2.5
Hospitality	0.0	2.9	1.0	0.0	2.5	0.0	0.0	9.8	0.0	0.0	3.2	2.0	0.0
General business	10.0	1.0	2.1	10.0	2.5	4.6	4.6	4.9	4.2	9.8	4.0	5.0	4.4
No coded connections	25.0	16.2	8.3	20.0	10.0	16.9	13.9	17.1	9.2	5.9	15.2	20.0	8.1
Diversity index	0.774	0.511	0.222	0.736	0.176	0.590	0.385	0.000	0.110	0.439	0.554	0.881	0.144
Reporting requirements	95	96	75	91	81	76	80	70	64	91.5	64.5	70	75

State	MD	MO	NC	NM	NY	OH	OR	RI	TX	VA	WA	WI	Mean	Standard Deviation
Agriculture	2.1	24.7	25.0	21.4	2.0	8.2	21.7	2.0	18.0	12.0	15.3	21.2	12.6	7.9
Health care	14.9	11.1	10.0	8.6	21.3	8.2	13.3	15.0	17.3	7.0	14.3	10.1	11.8	4.3
Education	24.1	19.8	16.7	27.1	32.0	12.2	8.3	26.0	18.0	16.0	27.6	14.1	21.0	7.1
Law	14.2	8.6	12.5	4.3	21.3	14.3	10.0	21.0	34.0	39.0	7.1	6.1	15.1	8.6
Finance and insurance	14.2	16.0	18.3	18.6	16.0	28.6	18.3	6.0	32.7	31.0	15.3	10.1	18.6	6.5
Real estate	17.7	30.2	34.2	35.7	21.3	25.5	53.3	11.0	32.0	42.0	23.5	22.2	30.9	12.3
Transportation	1.4	6.8	2.5	8.6	2.7	5.1	1.7	5.0	5.3	6.0	7.1	3.0	5.2	3.6
Mining	1.4	1.2	0.0	8.6	0.7	3.1	0.0	0.0	9.3	4.0	3.1	0.0	2.6	3.0
Communications and electronics	5.0	4.9	5.8	0.0	3.3	4.1	8.3	3.0	11.3	7.0	6.1	5.1	5.0	2.5
Construction	3.5	5.6	7.5	5.7	1.3	1.0	6.7	1.0	5.3	6.0	2.0	6.1	5.2	3.5
Hospitality	0.7	0.0	2.5	0.0	0.7	1.0	0.0	0.0	0.7	1.0	0.0	1.0	5.9	2.7
General business	7.1	6.2	4.2	5.7	7.3	7.1	8.3	11.0	8.7	3.0	6.1	5.1	28.2	5.3
No coded connxns	14.2	12.3	20.0	18.6	11.3	17.3	33.3	14.0	19.3	23.0	15.3	20.2	20.1	7.9
Diversity index	0.191	0.465	0.544	0.570	0.331	0.319	0.569	0.075	1.000	0.797	0.436	0.252	0.4	0.3
Reporting requirements	71	76.5	82.5	61.5	85	66	82	80	88	85.5	98	88	79.7	10.5

Primary Independent Variables

My first primary independent variable is employment for each sector. I measure this using U.S. Census data on employment for the employed civilian population over 16.9 The matches between industries as coded in the Center's data and as coded by the Census Bureau can be imperfect. Table 2 describes the Census categories used for each financial connection.

Table 2. Financial Connection Categories and Census Employment Share Categories

Financial Connection	Employment Share
Agriculture	Agriculture, forestry, fishing, and hunting
Health care	Health care and social assistance
Education	Education
Law	Law (ABA data)
Transportation	Transportation and warehousing
Mining and petroleum extraction	Mining
Communications and electronics	Information
Construction	Construction
Hospitality	Arts, entertainment, recreation,
	accommodation, and food services
Finance and insurance	Finance and insurance
Real estate	Real estate and rental and leasing
General business	Manufacturing, wholesale trade, and retail trade
No coded connections	N/A
Connection diversity	Reciprocal of Herfindahl index over employment shares

My second primary independent variables are professionalization and its components of compensation, session length, and staff support. I measure professionalization with Squire's (2000) index for 1995 and its components.

Control Variables

While the partisan split of the chamber is a part of my simple model of financial connections, I treat it as a control because I have no firm expectations about which industries should be correlated with which party, and to what degree. I measure the partisan split of the chamber with its percent Democratic. The models of financial connections also include three pure controls. I include the Center's rating of each chamber's stringency of reporting or openness. In states that require more connections to be reported, more connections to any given industry are likely to appear. While a simple control does not really merit a separate hypothesis, stringency should be positively associated with nearly every connection category, positively associated with the diversity of connections, and negatively associated with the proportion of legislators who have no coded connections. Following Squire (1992) and to help my results be more easily comparable to his, I also include a dummy for southern states and a Sullivan diversity index (Sullivan, 1973; Morgan and Wilson, 1990), updated with data from 2000.

Empirical Analysis

The Prevalence of Connections

I model the prevalence of connections to a given industry as a function of the employment share of that industry, professionalization or its components, the chamber's percent Democratic, stringency of reporting, state diversity, and region.10 As noted earlier, having more legislators connected to one sector does not require that there be fewer legislators connected to other sectors, but they do turn out to be correlated. Further, most of the independent variables are identical across models. It is therefore likely that errors will be correlated across models, and in fact they are ( in a Breusch-Pagan test). Accordingly, I run the models as seemingly unrelated regressions.11

Table 3 presents the results of regressions predicting connection frequency using the full Squire index. Hypothesis 1 is well supported: employment share is positively associated with the prevalence of the relevant connection nine of the 12 models where it appears. Employment share is not significant for education, health care, and finance and insurance. For education and health care, the null finding may reflect a low variance of employment shares; in the data set education employment varies only between 7.2 and 10.9 percent, and health care between 9.2 and 13.9 percent. The effects of employment shares are substantively strong. For example, increasing the agriculture employment share by one standard deviation (0.92) is associated with about half a standard deviation (3.5 percentage points) more legislators connected to agriculture. Law displays similar results; a one standard deviation increase in legal employment is associated with about a 1.2 standard deviation increase in legislators connected to law firms.

Table 3. SUR Models of Connection Prevalence, Full Squire Index

				Finance and	Real	No Coded
	Agriculture	Education	Law Firms	Insurance	Estate	Connections
Employment share	3.800***	0.919	14.766***	−0.339	9.095***
	(0.836)	(0.788)	(3.623)	(0.243)	(1.445)
Professionalization	−13.323	−21.918*	−14.014	−12.242	−7.157	25.427**
	(11.133)	(11.494)	(14.362)	(11.262)	(18.527)	(11.511)
Percent Democratic	−0.149**	0.186**	−0.015	−0.131**	−0.236**	0.180***
	(0.070)	(0.076)	(0.084)	(0.059)	(0.114)	(0.052)
South	2.262	−2.928	8.743***	4.156	−1.995	2.901
	(2.612)	(2.732)	(3.392)	(2.667)	(4.342)	(2.741)
Reporting requirements	−0.079	0.173	0.094	0.067	0.432**	0.356***
	(0.106)	(0.109)	(0.138)	(0.109)	(0.178)	(0.112)
State diversity	−2.555	23.423	71.104	13.641	−122.230*	11.047
	(43.651)	(43.811)	(55.015)	(42.203)	(71.035)	(42.603)
Constant	24.229	−15.601	−30.607	18.142	45.059	27.865
	(19.515)	(19.129)	(24.127)	(18.836)	(30.900)	(19.321)

	Health Care	Transportation	Mining	Communications and Electronics	Construction	Hospitality	General Business
Note Table reports coefficients from seemingly unrelated regression models. Standard errors in parentheses. *p ≤ 0.10; ** p ≤ 0.05; *** p ≤ 0.01, all two-tailed.
Employment share	0.919	2.827***	3.482***	1.664***	1.513***	0.439***	0.001
	(0.788)	(0.365)	(0.293)	(0.333)	(0.284)	(0.145)	(0.001)
Professionalization	−21.918*	−13.482***	−0.713	−2.238	6.305	−7.131*	−0.164*
	(11.494)	(4.233)	(3.336)	(3.846)	(5.684)	(4.074)	(0.093)
Percent Democratic	0.186**	−0.021	−0.022	−0.032	0.008	−0.016	−0.002***
	(0.076)	(0.027)	(0.022)	(0.023)	(0.034)	(0.025)	(0.001)
South	−2.928	−0.784	0.631	0.381	−2.014	−2.435***	−0.023
	(2.732)	(0.978)	(0.766)	(0.901)	(1.328)	(0.948)	(0.022)
Reporting requirements	0.173	0.032	0.006	0.109***	0.088*	0.083**	0.002*
	(0.109)	(0.041)	(0.031)	(0.037)	(0.054)	(0.039)	(0.001)
State diversity	23.423	11.596	21.385*	−11.854	−47.165**	24.361	−0.037
	(43.811)	(16.323)	(12.422)	(14.625)	(20.911)	(15.598)	(0.358)
Constant	−15.601	−9.965	−10.388*	−1.534	6.041	−11.480*	0.155
	(19.129)	(7.397)	(5.458)	(6.382)	(9.473)	(6.688)	(0.168)

Hypothesis 2 on professionalization fares less well. Its coefficient is statistically significant at the 0.05 level for only two industries, and at the 0.10 level for four more. These weak findings support Squire's (1992) original findings. Professionalization is not significantly associated with connections to agriculture or law, two industries among those advantaged in unprofessionalized legislatures. Professionalization is also positively associated at the 0.05 level with legislators who have no coded connections to any sectors, but it is important to remember that legislators with no coded connections need not be full-time legislators.12 Where it is statistically significant, professionalization is substantively important. Increasing professionalization by a standard deviation (0.133) is associated with 1.8 percentage points fewer transportation-connected legislators and 3.4 percentage points more legislators without coded connections, about half a standard deviation change in those dependent variables.

Among the control variables, reporting requirements are significant in seven models, with stricter requirements associated with more connections to each of the sectors and fewer legislators with no coded connections. State socioeconomic diversity and a South dummy were only rarely active. The percent Democratic is a significant predictor in seven of the 13 models.

Table 4 presents the results of regressions where the 1995–1996 Squire index has been broken into its components of pay, session length, and staff support. Pay is in thousands of dollars, session length in hundreds of days, and staff support in thousands of staffers. Squire's 1992 or 2007 articles provide more thorough discussions of the components (Squire, 1992, 2007). Overall, the results for variables other than professionalization remain similar, with two exceptions. First, the results for connections to health care change strongly when the Squire index is broken into its components, such that the coefficient on health and social services employment is now negative and significant. The second substantial change is that the effect of state diversity varies strongly, but is still generally not significant.

Table 4. SUR Models of Connection Prevalence, Squire Index Components

				Finance and	Real	No Coded
	Agriculture	Education	Law Firms	Insurance	Estate	Connections
Employment share	3.906***	0.536	16.532***	−0.283	4.735***
	(0.755)	(0.710)	(3.591)	(0.175)	(1.558)
Pay ($K)	−0.272***	−0.132	−0.013	−0.019	0.025	0.278***
	(0.089)	(0.101)	(0.132)	(0.101)	(0.166)	(0.097)
Session length (100 days)	5.630**	−2.992	−2.634	−4.033	−5.951	−0.069
	(2.582)	(2.875)	(3.817)	(2.926)	(4.882)	(2.800)
Staff (K)	1.665	2.872	−1.848	1.137	−0.313	−3.034
	(1.958)	(2.172)	(2.877)	(2.166)	(3.559)	(2.076)
Percent Democratic	−0.135**	0.235***	−0.079	−0.161***	−0.334***	0.184***
	(0.063)	(0.074)	(0.080)	(0.047)	(0.106)	(0.049)
South	1.382	−7.171**	9.962**	2.081	−2.839	6.634**
	(2.991)	(3.389)	(4.420)	(3.331)	(5.438)	(3.191)
Reporting requirements	−0.127	0.136	0.101	0.062	0.495***	−0.311***
	(0.092)	(0.102)	(0.136)	(0.105)	(0.170)	(0.100)
State diversity	−25.998	−40.302	90.960	−9.340	−71.734	60.389
	(46.815)	(50.380)	(64.979)	(48.854)	(83.042)	(46.955)
Constant	34.383*	13.884	−36.467	29.973	35.859	4.467
	(20.468)	(22.005)	(28.412)	(21.673)	(35.638)	(20.777)

	Health			Communications			General
	Care	Transportation	Mining	and Electronics	Construction	Hospitality	Business
Note Table reports coefficients from seemingly unrelated regression models. Standard errors in parentheses. *p ≤ 0.10; ** p ≤ 0.05; *** p ≤ 0.01, all two-tailed.
Employment share	−0.889***	3.472***	3.769***	2.164***	1.812***	0.491***	0.001
	(0.755)	(0.358)	(0.195)	(0.313)	(0.260)	(0.134)	(0.001)
Pay ($K)	−0.166***	−0.088***	0.030	−0.069**	−0.038	−0.082**	−0.002*
	(0.089)	(0.034)	(0.030)	(0.034)	(0.051)	(0.036)	(0.001)
Session length (100 days)	2.265**	0.830	−0.557	1.270	1.698	0.640	0.010
	(2.582)	(0.990)	(0.861)	(0.970)	(1.486)	(1.059)	(0.026)
Staff (K)	5.602***	−1.612**	−0.542	0.389	1.193	0.787	0.004
	(1.958)	(0.782)	(0.649)	(0.728)	(1.105)	(0.782)	(0.018)
Percent Democratic	0.187***	−0.023	−0.025	−0.032	0.006	−0.007	−0.002***
	(0.063)	(0.024)	(0.021)	(0.023)	(0.035)	(0.026)	(0.001)
South	−7.323***	0.718	1.135	0.110	−3.000*	−3.235***	−0.029
	(2.991)	(1.177)	(0.991)	(1.109)	(1.698)	(1.171)	(0.028)
Reporting requirements	0.076*	0.005	0.008	0.102***	0.087*	0.065*	0.001
	(0.092)	(0.036)	(0.031)	(0.034)	(0.052)	(0.037)	(0.001)
State diversity	−57.951***	35.097*	28.116*	−17.760	−51.886**	8.072	−0.173
	(46.815)	(18.645)	(15.059)	(16.784)	(25.448)	(18.515)	(0.421)
Constant	28.303***	−20.836**	−13.571**	−0.122	6.113	−4.828	0.231
	(20.468)	(8.343)	(6.496)	(7.251)	(11.152)	(7.724)	(0.192)

More important, however, is the effect of the professionalization components, which have wider effects than does the full index. Legislative compensation has a particularly sharp role. It has a 0.05 significant coefficient for six sectors, and one more 0.10 significant coefficient. On the one hand, the findings here support Hypothesis 3; increased pay is associated with fewer agriculture-connected legislators. However, it is not associated with fewer legislators connected to law firms, insurance, or real estate. That compensation is significant for so many industries supports the idea that the costs and benefits of service affect legislative recruitment. The effects of legislative compensation persist when session length, staff support, or both are dropped. Substantively, the effect of legislative compensation is difficult to characterize because pay varies widely. In my sample of states, compensation ranges from $300 in Alabama and Rhode Island to $72,000 in California with a standard deviation of $18,014. On the one hand, a one standard deviation increase in pay is associated with changes in connection prevalence between 44 and 70 percent of a standard deviation (1.3–4.4 percentage points). On the other hand, for those sectors where pay was statistically significant changing the prevalence of a connection by a standard deviation would require changing compensation by $25,000 to $40,000. The effect of session length is far less robust. It is rarely significant and when it is, it takes an unexpected positive sign. Further, while the effects of compensation are generally consistent even when it appears alone, the effects of session length are not. In sum, session length seems to have inconsistent effects.

The Diversity of Connections

Table 5 reports the results for a connection diversity index. These were run in the same SUR runs with the connection-prevalence models. Where the connection-prevalence models include the employment share of the industry, the economic diversity models include the employment diversity of the state.13 And where the connection-prevalence models use percent Democratic, the economic diversity models use percent in the majority party.

Table 5. SUR Models of Legislators’ Economic Diversity

	Model 1		Model 2
	Coefficient	SE	Coefficient	SE
Note Table reports coefficients and SEs from SUR models. *p ≤ 0.10; ** p ≤ 0.05; *** p ≤ 0.01, all two-tailed.
Employment diversity	0.064***	0.018	0.111***	0.015
Professionalization	−0.728*	0.379
Pay ($K)			−0.010***	0.003
Session length (100 days)			0.092	0.922
Staff support (K)			0.102	0.067
Percent majority	−0.008***	0.001	−0.007***	0.001
South	−0.002	0.091	−0.096	0.103
Reporting requirements	0.008**	0.004	0.006**	0.003
State diversity	−0.369	1.373	−2.217	1.457
Constant	−0.129	0.652	0.150	0.684

In terms of sign and significance, the results for economic diversity are quite similar to those for the connection-prevalence models. Again, structural factors loom large. Employment diversity and majority size are powerful predictors of the sectoral diversity of legislators. And professionalization and especially pay are strongly associated with less diversity, as Squire (1992) found with occupations. Given that professionalization and its components were associated with fewer legislators connected to many industries, and more of only a few, it makes sense that professionalization and pay would act to reduce the economic diversity of legislatures. However, there are important differences in the substantive effects of variables between the connection-prevalence and diversity models. In the connection-prevalence models, professionalization and its components generally had smaller effects than employment share. But in the economic diversity models, the reverse is true. The predicted effect of a one standard deviation increase in professionalization (or in pay) is approximately twice as large as the predicted effect of a one standard deviation increase in employment diversity.

Industrial “Representation,” Population Characteristics, and Industry Pressure

In my models, the patterns of legislative connections are largely byproducts of the underlying economic composition of the state, professionalization or pay, and the partisan makeup of the chamber. The industries themselves are not actively doing anything to foster their “representation” in the chamber. However, active industry involvement does present itself as an obvious alternative or addition to my simple model. A state might have many legislators connected to farming not only because the state has strong farm employment, but also because its farmers have been successful at helping their colleagues join the statehouse. More success in placing colleagues in the legislature should be a result of three factors. First, better-off and economically larger industries can, in potential, offer more resources and material support to their candidates. Second, a better-organized industry with strong unions or active interest groups or both should be better able to translate economic clout into actual support of specific candidates by being better able to solve the usual collective action problems. Finally, industries more tightly coupled to state law, such as law, insurance, and education, have more to gain from placing their members in the statehouse and so have a stronger incentive to deploy resources behind their candidates.

Ideally, one could simply place variables for economic power, organizational strength, and incentives into my existing model, but two problems prevent this. Organizational strength and incentives face operationalization problems: How do we know how well-organized an industry is, across industries and across states? An industry may have a strong union or professional organization in one state, and a weak one in another…but a more active interest group in the second. A further problem is that a given industry may have multiple competing organizations representing different groups and often working against each other. Incentives are similarly difficult to measure. The second major problem lies in including measures of economic importance. Economic clout is easy to operationalize through data on the GDP share of a given industry in a given state, but GDP share and employment share are collinear. Consequently, both cannot be included in the same model.

As one way to probe the plausibility of this alternative, I considered different observable implications of a world in which legislative connections are (in part) a result of how strongly industries push to get their members elected. In such a world, we should see campaign donations from the industry's interest groups flowing disproportionately (but not exclusively) to connected legislators. Or rather, it is difficult to conclude that an industry is pushing its members into office if that industry does not offer any special support to its connected legislators.

As a test of this connection, I examined donations made by teachers’ unions as identified by the National Institute for Money in State Politics. Teachers surely have every incentive to pursue such a tactic because both primary/secondary and postsecondary education are strongly regulated by the state, and government is the largest employer in the industry. Teachers’ unions are usually regarded as powerful, well funded, and mobilized, giving educators the ability to assist connected legislators in reelection. For (a negative) example, see Moe's (2011) Special Interest. Therefore, they should be prime candidates for pushing their colleagues into the legislature; a best-case scenario. I coded whether each legislator in my data set received a donation from a teachers’ union as identified by the Institute in the 1998 election cycle (1999 in Virginia).14 Then, in each chamber, I ran a logit using the education connection and a party dummy to predict whether a legislator received a donation. The results from these models appear in Table 6.

Table 6. Logit Models of Teacher-Union Donations to Sitting Legislators

	Connection		Democrat		Constant
State	Coefficient	SE	Coefficient	SE	Coefficient	SE	N
Note Table reports results from 25 separate logit models.In RI, no Republican received a donation and the model is for Democrats only. *p ≤ 0.10; **p ≤ 0.05; ***p ≤ 0.01, all two-tailed.
AK	−0.304	(1.321)	2.375*	(1.218)	−3.209***	(1.021)	40	0.083
AL	−1.133*	(0.690)	1.311*	(0.684)	1.654***	(0.452)	105	0.086
AR	−0.338	(0.636)	1.700**	(0.783)	−2.551***	(0.738)	96	0.035
AZ	1.288*	(0.683)	1.571***	(0.609)	−1.277***	(0.409)	60	0.003
CA	−0.234	(1.274)	4.283***	(0.851)	−1.126***	(0.412)	80	0.000
CO	−0.295	(0.776)	3.413***	(0.732)	−2.137***	(0.547)	65	0.000
CT	1.871***	(0.680)	2.932***	(0.459)	−1.808***	(0.392)	151	0.000
DE	1.701	(1.134)	0.372	(0.703)	0.162	(0.443)	41	0.210
FL	1.708**	(0.760)	2.217***	(0.449)	−1.514***	(0.309)	119	0.000
HI	−0.256	(0.667)	0.195	(0.709)	0.912	(0.660)	51	0.886
KS	1.445**	(0.639)	4.336***	(0.781)	−1.399***	(0.310)	125	0.000
KY	0.169	(0.490)	1.881***	(0.472)	−1.116***	(0.420)	100	0.000
MA	−0.011	(0.406)	1.939***	(0.636)	−2.119***	(0.614)	160	0.001
MD	0.920*	(0.457)	1.852***	(0.460)	−1.435***	(0.427)	141	0.000
MO	0.031	(0.553)	4.366***	(1.030)	−4.325***	(1.015)	162	0.000
NC	−0.657	(0.618)	1.100**	(0.454)	−1.573***	(0.368)	120	0.032
NM	−0.421	(0.653)	1.790**	(0.711)	−2.149***	(0.612)	70	0.020
NY	0.135	(0.445)	0.301	(0.415)	1.117***	(0.334)	150	0.711
OH	0.698	(0.830)	0.196	(0.476)	0.767***	(0.286)	98	0.546
OR	−0.385	(1.752)	4.156***	(1.137)	−0.906**	(0.377)	60	0.000
RI	1.280**	(0.520)			−1.367***	0.311	87	0.014
TX	0.142	(0.692)	4.384***	(0.652)	−1.194***	(0.296)	150	0.000
VA	0.385	(0.629)	1.432***	(0.503)	−1.893***	(0.432)	100	0.010
WA	0.150	(0.722)	4.132***	(0.644)	−2.004***	(0.470)	98	0.000
WI	1.962**	(0.861)	2.191***	(0.487)	−1.175***	(0.331)	99	0.000

The results of the models are mixed, but generally do not support the idea that educators are pushing educators into the legislature. The coefficient on an education connection was positive and significant in only seven states. In the remaining 18 states, the null result was far from subtle. With the exception of Delaware, the signs were in the wrong direction, or the standard error was larger than the coefficient, or both. At the same time, a minority of states do show the posited relationship. However, states with a positive and significant coefficient on connections had no higher or lower employment or GDP share for education than did other states. Further, educators in states with a positive and significant coefficient on connection did not “achieve” any more representation in their legislatures than did educators elsewhere, nor were they better able to translate employment or GDP shares into legislative seats in an interactive model. In sum, I looked at a sector with a very strong stake in legislative activity, with generally strong but variable organizational strength, and whose members should be easily mobilized. But even here, in 72 percent of states I examined, teachers’ unions did not target their campaign donations at connected legislators. The findings here make it difficult to accept that, as a general rule, active industry pressure is an important part of a legislature's economic background. They do, at the same time, support the idea that it can be an intriguing exception to the rule that bears further analysis.

Discussion and Conclusions

The distribution of legislators’ economic connections and the diversity of those connections have obvious potential consequences for which interests are favored in the legislative process. However, the work that examines these attachments has either been descriptive or has been structured around testing the effect of a very few theoretically key variables with a small set of controls. Consequently, our models of legislators’ economic attachments have failed to include the state's economic makeup. In this article, the most consistent predictor of the prevalence of a sector in legislature is the prevalence of that same sector among the state's citizens. Other variables that are consistently significant are the legislature's percent Democratic and legislative pay.

Three clear points emerge from my analysis. The first is that there is an element, but only an element, of economic descriptive representation among state legislatures. Economically, state legislatures remain distinct from their state populations. For example, every legislature has far more legislators connected to the legal world than the state as a whole does, and almost every state legislature is more strongly connected to agriculture and education than its citizens are. This lack of descriptive representation has sometimes been seen as problematic. What the results in this article show is that even though legislatures start from a less than representative position, the variation among them generally mirrors the variation among the states’ populations. States whose populations rely more strongly on agriculture or mining see that emphasis reflected, albeit imperfectly, in their legislators. This finding should gladden proponents of descriptive representation, as it shows that there are descriptive-representation forces at work in determining the economic makeup of legislatures.

The second main point that emerges, however, would be disheartening for reformers seeking to alter the mix of economic backgrounds state legislators are drawn from, irrespective of whether such changes are normatively desirable. My findings show that increasing the economic diversity of American legislatures, or changing the prevalence of any single industry, would be very difficult. Only employment, pay, and the partisan split were consistently able to predict the prevalence of industries. The employment patterns of states lie outside the control of reformers. Chambers with closer partisan splits are more diverse than chambers with heavy majorities, but intentionally altering the partisan distribution of the legislature would be difficult to achieve and have a vast array of potential side effects. While pay has clear effects on some connections, it has distinct limits as a lever on economic connections and diversity. Most obviously, effecting any serious change in the prevalence of a given sector would require very large changes in compensation, and the political environment seems unfavorable to large increases in legislative pay (see Kousser, 2005; Rosenthal, 1998, 2004). Second, while increasing pay would decrease the proportions of legislators connected to several sectors, the resulting shifts are predicted to actually decrease the diversity of the legislature. Rather than acting to increase the diversity of legislators’ economic connections, higher pay has the effect of simply reducing those connections.

A third point is that the effects of professionalization are heavily concentrated in its pay component. Indeed, this concentration is strong enough that compensation is frequently significant when the full Squire index is not, and at far higher levels of statistical significance. My results with the full Squire index were not too dissimilar from Squire's original results with occupational data: generally mild for specific industries.15 In contrast, I found a substantial number of statistically significant associations between connection prevalence and compensation. One possible interpretation of these findings is that while pay, staff support, and session length can clearly be combined into a meaningful and useful index, as the long history of success the Squire index has had demonstrates, in this particular circumstance, legislators and potential legislators face a simpler decision calculus. It ought not be surprising that serious career, work-life, and family welfare decisions are strongly influenced by the net financial costs imposed by legislative service.

1. 1

   This should be unsurprising as Maddox (2004a) used a precursor of the source data I use.

2. 2

   For example, Wahlke et al. interviewed a few legislators who admitted that they had pursued legislative service in order to push for laws that would help them or their industry (1962:117–19).

3. 3

   Or, if there are biases in the acquisition of interest, a pool of potential candidates that covaries with population demographics.

4. 4

   At least so long as the session does not interfere with work-heavy times of the year.

5. 5

   The miscellaneous business category includes a variety of essentially retail sectors as well as steel, chemicals, textiles, and miscellaneous manufacturing, but excludes manufacturing and related activity in other coded sectors. Automobile and aircraft manufacturing, or their parts, are coded as transportation.

6. 6

   Another approach would be to divide the proportion of legislators connected to each industry by the sum of those proportions to create the relative density of each sector, and then generate a Herfindahl index from those data. Analyses using that as a dependent variable yielded similar results.

7. 7

   The education category includes also legislators with a “government” connection where the interest's unedited name includes “education,” “school,” “college,” “university,” or “teacher.”

8. 8

   Except for property rental.

9. 9

   Question P49 in SF3. The one exception is employment in law firms. For this variable, I divided the total number of attorneys in each state from American Bar Association data by the total employed civilian population over 16 from Census SF3 data. While this variable misses nonattorney employment in law firms, it should be a good proxy for total legal employment.

10. 10

    A full replication data set is available on request.

11. 11

    Models were run using the “sureg” package for Stata, which does not include a switch for Huber-White standard errors. Run as a set of independent models for each industry, only real estate and construction were heteroskedastic. Independent models with robust SEs did not strongly differ from the seemingly unrelated regression (SUR) results I report except for real estate, where the coefficient on employment shares was no longer significant (two-tailed p = 0.906).

12. 12

    In an initial exploration of five nonrandom chambers, 30 percent of legislators reporting their occupation as “Legislator” had no coded connections, but 30 percent each were connected to miscellaneous business or finance, insurance, and real estate.

13. 13

    The reciprocal of a Herfindahl index calculated over sector shares of employment using the full array of Census SF3 data.

14. 14

    In Arkansas and Delaware, teachers’ unions did not make any donations to lower chamber candidates in 1998, so I substituted data from 2000.

15. 15

    Though I found 0.05 significant results for transportation.