Interest Groups & Advocacy

, Volume 7, Issue 3, pp 258–288 | Cite as

Financing the education policy discourse: philanthropic funders as entrepreneurs in policy networks

  • Sarah Reckhow
  • Megan Tompkins-Stange
Original Article


We examine the spread and influence of ideas supported by philanthropic foundations within the context of a broader policy network. Our case focuses on the development of policy related to teacher quality—a field involving academic research, think tank involvement, and interest group participation. We conduct discourse network analysis of testimony from 175 Congressional hearings from 2003 to 2015 to examine network ties based on shared policy preferences expressed in hearings, which were used to create networks linking policy actors via shared policy preferences. We also conducted 51 interviews with funders, grantees, and policymakers involved in the policy debate over teacher quality. We examine the spread of a key policy reform promoted by several large foundations, particularly the Gates Foundation: test score-based evaluation of teachers, with a focus on value-added evaluations. We show that expert witnesses in hearings who were funded by foundations shared policy preferences with regard to teacher evaluation at a statistically significant level, compared to non-grantees. We find that a group of major national foundations were sponsors of the advocacy groups that were central in Congressional hearings. We show that these funders were acting as policy entrepreneurs—strategically promoting the spread of favored ideas to encourage uptake by policymakers.


Policy networks Discourse analysis Philanthropy Education policy Mixed methods 


Philanthropic foundations have played central roles in shaping and influencing education policy throughout the last century. Historically, philanthropists have adopted what Clemens and Lee (2010, p. 62) describe as “an expert-led model of change,” funding empirical research in order to influence policy priorities, while relying on impartial intermediary actors to disseminate best practices within the field. The past two decades, however, have seen a shift away from this approach as a number of large foundations have elected to engage more directly with the political process by supporting advocacy organizations, investing in the implementation of major policy reforms, and ensuring that their research investments are targeted to advance specific policy priorities (Reckhow 2013; Tompkins-Stange 2016).

As foundations have adopted a more advocacy-oriented approach to grantmaking, their sponsorship of research has also changed. From 2000 to 2010, major K-12 education funders gave a smaller share of grant dollars to universities, while providing a larger share of funds to national advocacy organizations, think tanks, and intermediaries (Reckhow and Snyder 2013). These growing investments in explicitly policy-oriented research and advocacy organizations suggest that foundations are attempting to shape how policymakers consider ideas. However, studies have yet to empirically quantify the relationship between foundation support and the uptake of specific policy ideas, as existing research predominantly draws from surveys and qualitative interviews. This study extends prior work by using original data and mixed methods research in order to examine the spread and influence of ideas supported by philanthropic foundations, within the context of a policy network centered on teacher quality.

Teacher quality has been politically contested terrain for decades, as policymakers and advocates have debated issues from pay for performance, to unionization, to gender equity, to training and professionalization (Goldstein 2014). Since the passage of No Child Left Behind in 2001, however, these debates have taken on new depth and importance. The increased sophistication of state-level data systems to collect and analyze student data gathered through mandated state testing has enabled the application of econometric methods to estimate causal relationships between teachers’ instruction and individual students’ testing outcomes, as opposed to aggregated data at the school level. In the past decade, policy discourse has converged around the need to identify, assess and support high-quality teachers while removing low-quality ones, as indicated by student learning outcomes measured by standardized tests. Federal policies under the Obama administration, such as Race to the Top and the No Child Left Behind waiver program required states to create systems to evaluate teachers.

Traditionally, teachers—most prominently represented by teachers’ unions—have been among the most dominant interest groups in education politics (Moe 2011). Yet despite widespread teacher opposition to test score-based evaluation systems, this idea gained support among a broad swath of policymakers from both political parties (Galey et al. 2017). How did teacher evaluation become the dominant idea for addressing teacher quality and improving education in the United States?

To understand how the ideas shaping teacher quality policy have evolved since No Child Left Behind, this study unpacks the policy debate by analyzing witness testimony in Congressional hearings on teacher quality between 2003 and 2015, comparing the policy preferences of foundation grantees with those of non-grantees. Further, the study interrogates these findings with interviews with key witnesses, policymakers, and funders. Thus, within a complex policy debate occupied by hundreds of actors and organizations, we are also able to address the following question: how did philanthropic funders promote a specific set of ideas to shape the policy debate?

The study employs a mixed method research design using an innovative new methodological technique, discourse network analysis, that links content analysis to social network analysis, providing a way to combine the study of actor relationships with the content and sources of their policy beliefs. This approach illuminates how policy preferences are shared via networked relationships (Fisher et al. 2013). These relationships are contextualized and extended in a more granular fashion through 51 in-depth interviews with key actors, who were identified through social network analysis, including funders, researchers, advocates, and policy elites.

Through these complementary research methods, the study empirically traces the trajectory of political discourse around teacher quality over a 12-year period, illustrating how certain ideas related to teacher quality—particularly with regard to assessment and evaluation—entered, diffused, and later dominated policy debates. This analysis shows which ideas were promoted and shared by foundation grantees in policy debates, as indicated by their statements in Congressional hearings, while our interviews demonstrate the role of funders as entrepreneurs and significant strategic coordinating forces among the key actors in the network.

Foundations as entrepreneurs within policy networks

Actors within informal policy networks are embedded in an information-rich system—where a wide range of actors promote often-conflicting information about policy problems and solutions (Kingdon 2011; Rhodes 2006; Baumgartner and Jones 2012). Within these networks, actors receive and transmit policy ideas, innovations, and solutions, but because of cognitive and institutional limitations, actors cannot devote attention to many ideas at once—a phenomenon known as “serial processing,” or “selective attention processing” (Jones and Baumgartner 2005). Thus, they rely on others within the network to provide signals about the legitimacy or efficacy of various ideas in order to shape policy outputs.

Within the network, influential actors assume the role of policy entrepreneurs (Bushouse 2009; Quinn et al. 2014). Kingdon (2011, p. 12) defines policy entrepreneurs as “…people who are willing to invest their resources in pushing their pet proposals or problems,” and who are responsible “for coupling solutions to problems and for coupling both problems and solutions to politics.” The fundamental defining characteristic of the policy entrepreneur is what Hwang and Powell (2005, p. 23) term an “empowered actorhood”—the ability to champion policy solutions throughout the political process. Thus, policy entrepreneurs go beyond the role of brokers that adjudicate between approaches and highlight expedient solutions, to more central, hands-on, and assertive leadership in the network.

The study examines how philanthropic foundations have operated as policy entrepreneurs in federal education policy debates—specifically, on teacher quality. Other researchers in education policy have demonstrated how funders play the role of a “hub” for distributing information to organizations that advocate to market-based reforms (Scott and Jabbar 2014). Foundations that fund a wide variety of interest groups can catalyze a field around a common issue position, by both igniting a new area of focus and amplifying an existing area through strategic investments. For example, Reckhow and Tompkins-Stange (2015) found that the Gates and Broad Foundations made targeted grants to organizations that promoted a common set of policy priorities around teacher quality, thus ensuring that multiple participants in the policy debate echoed the same idea—conferring legitimacy upon it and diffusing it throughout the network. Additionally, entrepreneurial funders can lay the groundwork for effective policy implementation through capacity building—by funding demonstration projects and supporting the initial development of technical infrastructure within state or local bureaucracies, as Bushouse (2009) shows in the case of the Pew Foundation’s investments in early childhood education.

If foundations do indeed behave as policy entrepreneurs within a teacher quality policy network, we would expect to see evidence of two indicators in the study’s findings. First, we would observe a high degree of shared policy preferences among organizations that are foundation grantees. Second, we would expect interview subjects to discuss funders as the originators or catalysts in promoting specific ideas across organizations and creating conditions to enable future implementation.

The study thus explores the following empirical questions in order to examine whether and how foundations might operate as policy entrepreneurs on teacher quality issues:
  • Do hearing witnesses who received major foundation grant funding differ systematically in the ideas they share in testimony compared to witnesses who are not grantees? Are foundation grantee witnesses more likely to support preferences related to teacher evaluation, controlling for other factors?

  • How do actors involved in federal education policy explain their relationships to, and the roles of, major foundations in the development of ideas?

Research design and methodology

Congressional hearings are an important formal venue for communicating policy information, and information provided in Congressional testimony about policy effectiveness is positively associated with proposal enactment (Burstein and Hirsh 2007). Actors who deliver testimony in hearings have the opportunity to provide valuable information that policymakers could act upon and adopt as new preferences, but receptiveness to new information is likely to depend upon existing preferences and prior agreement with organizations participating in the hearing.

This study is based on analysis of 551 witnesses from 175 Congressional hearings from 2003 to 2015 on the topic of teacher quality (excluding members of Congress, as we focus solely on the statements made by witnesses). The hearings were downloaded from the US Government Printing Office website. The witness testimony was content-analyzed to construct discourse networks based on shared policy preferences. These data were also analyzed using logistic regressions and relational contingency tables (a technique for analyzing network data) and is further explored through interviews with key actors related to the issues discussed in the hearings, including policy staff, researchers, interest group advocates and funders.

Discourse network analysis links social network analysis to content analysis, providing a way to combine the study of actor relationships with the content and sources of their policy beliefs (Leifeld 2013). Studies drawing on the Advocacy Coalition Framework have demonstrated the strong link between policy beliefs and actor alliances (Weible and Sabatier 2014; Ansell et al. 2009). These networks are not static, however, and discourse network analysis can show how networks evolve over time, as well as the shifting dynamics between and within advocacy coalitions (Fisher et al. 2013). In addition, discourse network analysis can be employed to examine the link between the structural locations of network actors and their policy-oriented activities.

A discourse network is constructed by analyzing actors’ attitudes expressed in a public arena (e.g., national media, Congressional hearings) and creating ties between actors based on shared views (Leifeld 2013). For example, two actors who publicly state that teacher quality should be assessed with value-added models would share a link in a discourse network. As Leifeld observes, in prior studies of policy coalitions, “the actual processes of policy learning and policy change largely remain a black box” (2013, p. 171). Discourse network analysis provides a technique to unlock this black box, and to illuminate the mechanisms within, by examining the emergence of beliefs within a network and the relationships that develop around shared beliefs, or what Leifeld calls the “discursive layer” of politics (2013, p. 173).

The Discourse Network Analyzer software developed by Leifeld (2013) was used to content analyze the hearings and articles. A team of human coders read each Congressional hearing and coded statements using a specified set of policy belief categories (see “Appendix”). The results of this coding process were validated by a research supervisor. The research supervisor maintained the coherence between individual coders by cross-checking the work by each coder. There is no formal measure of intercoder reliability. When there was disagreement between a coder and the supervisor, we discussed the disagreement and sought a consensual solution. All spoken and written testimony provided in the hearings were coded, but the question and answer segments that typically follow testimony were not.1

To organize our analysis, we used a concept drawn from the Advocacy Coalition Framework (ACF) theory of the policy process (Jenkins-Smith et al. 2014), which explains how coalitions and network structures operate in high-conflict policy areas. A key element of the ACF is its emphasis on the beliefs that policy actors hold, which comprise three tiers: deep core beliefs, which represent broad normative values; policy core beliefs, which are relevant to a specific topical area of policy; and secondary beliefs, or policy preferences, which concern the technical strategies and instruments that may be used to achieve a desired outcome. Three deep core beliefs formed the basis of our codebook: efficiency, professional expertise, and equality. Our coding focused on identifying specific policy preferences related to the belief that teacher quality could be improved by addressing efficiency—an emphasis on economic cost-benefit and optimization of policy performance (Wood and Theobald 2003); equality—the use of political authority to redistribute critical resources (Kirst and Wirt 2009); and professionalism (Wilson 1991; Meier and O’Toole, 2006)—an emphasis on professional autonomy and education-specific knowledge shared by educators.

Policy preferences are not always siloed within a single core policy belief. For example, a policy preference for using differential compensation or “pay for performance” systems could be classified under the “Teachers must be evaluated and held accountable” core policy code under Efficiency or under the “Teachers are professionals” core policy code under Professional Expertise if the speaker mentions a high level of teacher involvement in developing the system. As described in more detail below, our analysis does not prioritize one code over another but instead employs multiple lenses to understand how some ideas gain traction and are diffused while others may not.

Table 1 illustrates how our data were coded within the deep core belief of efficiency (the full codebook is included in the “Appendix”).
Table 1

Sample codebook based on advocacy coalition framework

Tier 1: deep core belief (normative values)

Tier 2: policy core belief (specific to topical area)

Secondary belief (technical strategies)

We refer to secondary beliefs as policy preferences


Hold schools accountable for student performance

Maintain/establish a system of annual high-stakes testing


Use standardized testing to measure individual teacher inputs


Use school level testing to assess teacher quality


Teachers must be evaluated and held accountable

Use evaluation systems with growth models


Use evaluation systems with multiple measures


Use evaluation systems with student feedback


Use evaluation systems with classroom observations


Use evaluation systems with value-added models


Use evaluation systems with student reviews


Use evaluation systems with peer reviews

In parallel to our coding of policy preferences, we also coded each hearing witness on two dimensions: organizational characteristics and foundation funding. Witnesses’ organizational affiliations were identified and categorized by institutional type: universities, unions, school districts, and intermediary organizations (a broad category that includes think tanks, advocacy groups, and education nonprofits). Each organization was also classified by its status as a grantee of one of 17 major foundations involved in educational policy—particularly teacher quality.2 These funders were identified based on Reckhow (2013) and Reckhow and Tompkins-Stange’s (2015) methods for grant data analysis and coding research reports funded by philanthropies. Witnesses were categorized as “grantees” if their organizations had received funding during the year of the testimony or two years prior to it. We also coded witnesses that were supported by the Gates Foundation and the Broad Foundation, which our prior research identified as the two most active funders of teacher quality initiatives. In the teacher quality hearings from 2003 to 2015, 42% of the witnesses were coded as foundation grantees, including witnesses that represented organizations receiving funding from multiple foundations. Within the broader grantee category, 27% were Gates Foundation grantees, and 13% were Broad Foundation grantees. Based on other categories of witnesses that we coded, 15% of witnesses represented universities, 14% represented school districts, and 24% represented intermediary organizations. Only 2.5% represented unions, a remarkably low level for a body of hearings focused on a heavily unionized profession, but also a sign of the diminishing dominance of teachers unions in education politics.

In parallel with network analyses, 51 in-depth qualitative interviews were conducted with actors who are centrally involved in the teacher quality policy network. Interviewees were identified by triangulating data to construct a sample of key actors within the teacher quality policy network. A compiled database of potential informants included the expert witnesses who testified in Congressional hearings, the authors of major research reports on teacher quality that were cited in the hearings, leaders of key intermediary organizations involved in teacher quality policy advocacy, and media sources who wrote articles, speeches and white papers about teacher quality. Interviewees were recruited through personal contacts with orienting figures and by systematic recruitment procedures, including emails, calls, and face-to-face meetings.

Standardized interview protocols were tailored to interviewees based on their category of profession or expertise (i.e., researchers, advocates, policy staff, media, funders, thought leaders). The protocols were further refined to individual interviewees based on extensive background research on their work and publicly stated views, containing specific probes for each question depending on the informant’s position within an organization. The protocols were structured to align with standards for elite interviewing, as delineated by Dexter (1970), Berry (2002) and Goldstein (2002), allowing for a semi-structured, open-ended approach rather than a rigid adherence to a script. The results of the discourse network analysis were used iteratively to inform the lines of inquiry for the interviews, and protocols were refined as new relationships and indicators in the network data emerged. The study is designed using an inductive approach, wherein the interview process is fundamentally sensitive to new information that may expand or contradict working hypotheses.

Approximately half of the interviews were conducted in person and half via phone. Interviews were tape-recorded with the permission of the subjects, except in some situations where sources were reluctant to share information on tape, in which case the researchers took extensive notes, including verbatim quotes, and documented their impressions immediately after the conclusion of the interview. Recorded interviews were subsequently transcribed verbatim, averaging approximately 60 min in length, ranging from 25 to 100 min. Subjects were guaranteed anonymity and are referred to as “interviewees” or “sources” with general description of their professional roles, a choice that involved sacrificing some granular details at the expense of ensuring accurate portrayals of phenomena they described.

The interviews were analyzed in several stages. A group of transcribed interviews were selected at random to inductively develop and assign a first round of general codes that descriptively categorized elements of the data. This initial scheme developed into a baseline codebook, which was added to in an iterative fashion as new interviews were conducted, engaging in in vivo coding when new themes emerged from the data and also collapsing existing codes into broader categories that became apparent across the corpus of interviews.

Visualizing the policy networks

To begin the analysis, we examine the preferences of hearing witnesses visually using network diagrams. We represent these figures as two-mode networks—meaning they display relationships between hearing witnesses and the different policy preferences they supported based on our hearing coding. In particular, we are interested in examining how the preferences related to teacher evaluation and accountability are related to other types of preferences involving different deep core beliefs, such as professionalism and equality, and which actors were most supportive of different categories of preferences. For these visualizations, we have grouped specific policy preferences together within the policy core belief that these preferences support. In other words, all of the specific preferences related to teacher evaluation, such as: “Use evaluation systems with value-added models” are grouped with the overarching policy core belief: “Teachers must be evaluated and held accountable.” The same is true for all of the other preferences displayed in the networks.

Figure 1 represents the network of support for various policy beliefs from 2003–2015 among the witnesses coded as foundation grantees. The gray circles represent the policy core beliefs and the black circles represent the hearing witnesses. Larger gray circles indicate more witnesses support those policy core beliefs—meaning these have higher centrality. The policy core belief and preferences with the highest centrality in the grantee network are “Teachers must be evaluated and held accountable.” The category with the next highest centrality among grantee witnesses, “Teachers and education leaders respond to performance-based incentives.” The close proximity of these two categories in the network shows that there is overlap in support among the witnesses for these two sets of beliefs and preferences. In other words, many grantee witnesses who support evaluation reform also support performance-based incentives. Among the grantee witnesses, there is relatively less support for policies related to equality. For example, the category, “Variation in teacher quality is often associated with other forms of inequality in schooling,” ranks second to last in centrality among the grantee witnesses.
Fig. 1

Grantee policy core belief network: 2003–2015

Figure 2 represents the network of support for policy beliefs among the non-grantee witnesses for 2003–2015 hearings. The network figure is constructed in the same manner as Fig. 1. In this figure, the most central category is an efficiency belief related to economic competitiveness: “Teachers must prepare students to enter a modern competitive workforce.” This is closely associated with another highly central category among non-grantees related to professionalism beliefs—“Teachers must be prepared in schools of education with training tailored to their job.”
Fig. 2

Non-grantee policy core belief network: 2003–2015

Table 2 provides the raw figures for in-degree centrality of key policy beliefs from Figs. 1 and 2, offering a direct comparison of grantees and non-grantees. In-degree centrality shows the exact number of times any witness mentioned a preference within each belief category. As the table shows, efficiency beliefs related to policies like teacher evaluation rank the highest for grantees. These beliefs also receive support from a fairly large number of non-grantees based on centrality, although there are a much larger number of non-grantee witnesses in the hearings (there are 87 more non-grantee witnesses than grantees), so the percent of mentions among grantees is higher. In general, the non-grantees show support for policies falling across a wider spectrum of the three belief categories. Meanwhile, the largest gap in centrality between the grantees and non-grantees is related to equality beliefs, specifically: “Variation in teacher quality is often associated with other forms of inequality in schooling.” This has relatively high support among the non-grantees. By comparison, it is among the lowest ranked preferences for the grantees. These rankings are interesting descriptively, but they tell us little about whether witnesses who are grantees have distinct preferences controlling for other factors, nor can we discern whether a particular set of witnesses is more likely to share the same beliefs with other similar types of witnesses.
Table 2

Teacher quality preference in-degree centrality: grantees and non-grantees (based on networks in Figs. 1, 2)


Grantee network (231 actors)

Non-grantee network (318 actors)

Efficiency beliefs


 Teachers and educational leaders must be evaluated and held accountable



 Teachers and educational leaders respond to performance-based incentives



 Teachers must prepare students to enter a competitive, modern workforce



Professionalism beliefs


 Teachers must be prepared in schools of education with training tailored to their job



 Teachers build capacity and knowledge through collaborative professional development



Equality belief


 Variation in teacher quality is often associated with other forms of inequality in schooling



Policy preference analysis findings

The network diagrams suggest that grantees and non-grantees have somewhat different levels of support for various ideas to improve teacher quality—but do these differences persist even when controlling for other witness characteristics? We will assess this data in two ways—(1) logistic regressions, and (2) tests of network structure. For logistic regression analysis, we use data from testimony provided by all of the witnesses in our dataset. Thus, each unit of analysis is a statement or testimony by an individual participating in a Congressional hearing.

The models in Tables 3 and 4 are logistic regressions predicting whether a witness mentioned certain policy preferences in their testimony. We compare the results for two different sets of policy preferences—evaluation preferences (Table 3) and equality preferences (Table 4)—to assess whether distinct witness characteristics are better predictors of each set of preferences. The evaluation preference dependent variable includes all mentions of specific policy preferences under the policy belief category: “Teachers must be evaluated and held accountable.” The model predicts which witness characteristics are associated with testimony that includes these evaluation preferences. The equality preference dependent variable includes specific preferences aligned with three categories of policy beliefs: “Improving education requires a comprehensive approach”; “Variation in teacher quality is often associated with other forms of inequality in schooling”; and “Historically, the composition of the teacher workforce and pedagogical preparation has disadvantaged certain populations.” In each model, year is included as a variable to control for considerable annual variations in the number of hearings and attention to particular issues. In addition to year, we included four categories of hearing witness affiliations: universities, school districts, unions, and intermediaries. Lastly, we examine the relationship between witness preferences and foundation support, by comparing three different dummy variables: (1) witnesses representing organizations supported by any of 17 major funders associated with teacher quality; (2) witnesses representing organizations supported by the Gates or Broad Foundations; (3) witnesses representing groups supported by the Gates Foundation. (We separated Gates and Broad in the last model in order to account for the fact that Gates’ endowment, at $40.3 billion in 2016, is nearly 13 times larger than Broad’s, at approximately $3 billion.) Thus, each foundation related variable examines a narrower set of witnesses to assess whether general philanthropic support is associated with certain policy preferences, or whether there is a stronger relationship with specific funders.
Table 3

Predicting teacher evaluation preferences in congressional testimony, 2003–2015


Model 1

Model 2

Model 3


0.260** (0.047)

0.262** (0.047)

0.259** (0.047)


0.336 (0.319)

0.331 (0.315)

0.351 (0.314)


− 0.970* (0.434)

− 0.992* (0.432)

− 1.030* (0.433)

Foundation grantee

0.572* (0.259)

Gates/broad grantee

0.676** (0.263)

Gates grantee

0.759** (0.269)


− 0.658* (0.317)

− 0.706* (0.317)

− 0.693* (0.312)


0.392 (0.635)

0.319 (0.637)

0.315 (0.638)









Table entries are unstandardized regression coefficients with standard errors in parentheses. For a two-tailed test of significance, *p < 0.05; **p < 0.01

Table 4

Predicting equality preferences in congressional testimony, 2003–2015


Model 1

Model 2

Model 3


0.055 (0.123)

0.057 (0.035)

0.055 (0.036)


0.700* (0.278)

0.594* (0.275)

0.599* (0.274)


− 0.179 (0.314)

− 0.352 (0.303)

− 0.370 (0.303)

Foundation grantee

− 0.102 (0.211)

Gates/broad grantee


0.229 (0.217)

Gates grantee


0.286 (0.223)


0.064 (0.249)

− 0.114 (0.245)

− 0.116 (0.240)


1.157* (0.576)

1.122* (0.576)

1.122* (0.576)









Table entries are unstandardized regression coefficients with standard errors in parentheses. For a two-tailed test of significance, *p < 0.05; **p < 0.01

Table 3 shows the models predicting evaluation preferences with several characteristics of the hearing witnesses. In Model 1, the variable for foundation grantee is a positive and statistically significant predictor of evaluation preferences, while witnesses that represent universities or intermediary organizations are statistically significant and negatively associated with evaluation preferences. The negative result for the intermediary variable is surprising, given that many key intermediary organizations, such as Education Trust and The New Teacher Project, were vocal supporters of teacher evaluation reforms.

In Models 2 and 3, we use a different dummy variable for philanthropic funding. First, in Model 2, we include a dummy variable for Gates and Broad grantees, and in Model 3, we examine the results for Gates grantees. In these models, the results for the other independent variables do not change. Yet the grantee variables in Models 2 and 3 have larger coefficient sizes, suggesting that evaluation preferences may be more strongly associated with certain funders, especially the Gates Foundation. For example, using the results in Model 1, the predicted probability for a foundation grantee mentioning evaluation preferences is 0.64. Meanwhile, using the coefficient for Gates Foundation grantees in Model 3, the predicted probability for mentioned evaluation preferences grows a bit to 0.68. Thus, witnesses that are Gates Foundation grantees are slightly more associated with mentioning policy preferences supporting teacher evaluation.

The models in Table 4 predict equality preferences with characteristics of hearing witnesses. In Model 1 the variable for grantee is no longer statistically significant, similarly, Models 2 and 3 show that neither Gates/Broad grantees nor Gates grantees are significantly associated with equality preferences. Two other characteristics of hearing witnesses are positive and statistically significant in all three models—witnesses representing school districts and witnesses representing unions. These results are not surprising, particularly for unions—union leaders often voice support for redistributive policies and increased funding for schools and teachers. Based on the estimated coefficient for Model 1, the predicted probability of a union representative mentioning equality preferences is 0.76. Overall, we find that foundation grantees focus on a unique set of policy preferences, with a particular emphasis on teacher evaluation preferences. Moreover, witnesses who represent organizations supported by the Gates Foundation have the highest likelihood of discussing teacher evaluation preferences.

Network analysis findings

The logistic regression results show that grantee witnesses are generally more supportive of teacher evaluation policy ideas, controlling for other factors. Yet this type of analysis treats each individual hearing witness as an isolated observation, rather than considering the relational characteristics of the data. Hearing witnesses are not necessarily isolated individuals—they are participants in larger policy networks who communicate with other actors involved in related issues. Our data is also compiled as network data, and we can analyze whether specific actor categories, such as grantee and non-grantee, or Gates grantee and those who did not receive Gates funding, are associated with the relational structure of the data. For the network analysis, we focus on a specific segment of our preference network—the two categories of preferences with the highest centrality among the foundation grantees—(1) teacher evaluation and accountability preferences and (2) performance-based incentives.

To examine this network, we collapse the two-mode policy preference and actor network (visualized in Figs. 1, 2) into a one-mode network of ties directly between actors including both foundation grantee witnesses and non-grantees. In other words, a tie between any two actors in this network indicates that those actors both share preferences for one of the two previously mentioned policy preference categories (evaluation or incentives). We use this data to assess the following question: does witness affiliation as a foundation grantee (or Gates Foundation grantee) predict more shared preferences related to teacher evaluation and performance-based incentives? In this model, rather than predicting the preferences expressed by individuals, we are predicting the density of ties (shared beliefs) between pairs of witnesses. Thus, for this analysis we examine the entire network of witnesses who share the same preferences, in order to compare grantees and non-grantees, rather than examining each witness as an isolated actor.

To assess this question, we use a technique called relational contingency table analysis (Borgatti et al. 2002; Hanneman and Riddle 2005). This type of analysis examines the ratio of actual versus expected ties both within and between groups. In other words, it tests for homophily in the network: do actors who share a characteristic or attribute tend to share more network ties? In our study, we are interested in network homophily among the grantee witnesses. In order to predict the number of expected ties, the model generates 10,000 random permutations of the network—essentially a bootstrapped Chi-square analysis. These randomly generated networks are compared to the actual network data to determine the likelihood that the observed network would appear randomly, without the structure suggested by the homophily-based ties.

The results of our relational contingency table analysis show that there are a greater than expected number of ties among the foundation grantee witnesses—a ratio of 1.33 actual ties to expected ties. The opposite is true of non-grantee witnesses; the ratio of actual to expected ties in this group is 0.78. The Chi-square test for the model is statistically significant, p = 0.021. Additionally, we repeated the same analysis, but instead of foundation grantees generally, we look specifically at Gates Foundation grantees. The results for Gates grantees are stronger, with a ratio of 1.69 actual ties to expected ties. The Chi-square test for this model is also statistically significant, p = 0.002. This analysis shows that foundation grantee witnesses are significantly more likely to share preferences related to teacher evaluation and performance-based incentives than their non-grantee counterparts. Furthermore, the likelihood of shared preferences among Gates Foundation grantees is even stronger than foundation grantees generally.

Overall, our quantitative results show that witnesses representing organizations funded by the Gates Foundation are much more likely to support an aligned set of policy preferences related to teacher evaluation reform. We find that non-grantee witnesses tend to support different policy preferences, related to the policy core belief categories of professionalization and equality. While these results do not draw on direct observations of foundation behavior, they do demonstrate the observable consequences of foundation grantmaking patterns in the political realm. These findings are aligned with the patterns we would expect to observe if foundations are acting as entrepreneurs—the recipients of foundation funding (particularly Gates funding) are generally united in their publicly stated policy preferences.

Nonetheless, more direct evidence of foundation strategy and behavior is necessary to fully assess whether foundations are, in fact, the policy entrepreneurs behind these networks of shared policy preferences. While our quantitative findings show compelling associations between foundation-funded policy actors and expressed policy preferences, we cannot accurately assess causality using statistical analysis alone. To complement and extend these findings, we completed 51 in-depth qualitative interviews with elite actors in the teacher quality policy discourse network. This method of triangulation enables us to begin uncovering the specific mechanisms within the black box that foundations and advocates employed to advance their policy preferences.

Advancing teacher evaluation as a central policy priority

Grantees’ convergence around teacher evaluation as a dominant priority was also reflected in our findings from the qualitative interviews. Our sources shared a narrative about the education reform community’s emphasis on galvanizing political energy around teacher evaluation, because they viewed it as a baseline for the rest of a comprehensive education reform overhaul. Foundations were key policy entrepreneurs within this strategy, providing funding and access to support their reform priorities.

Interviewees shared that an elite group of peers representing foundations, their grantees, and sympathetic policy elites were strongly committed to institutionalizing teacher evaluation as a federal policy priority. They viewed evaluation as a cornerstone of their theory of change, built on the assumption that the existing teaching profession needed to be disrupted. To this group—which several interviewees described as the “echo chamber”—establishing teacher evaluation systems was a precondition for accomplishing other related goals, such as alternative teacher certification and differential compensation. One interviewee, an academic researcher, described this dynamic in the following way:

Teacher evaluation [became] almost the thing that is needed to do anything else. If you don’t have a good evaluation system you can’t look seriously at the progress of your education preparation institutions, you can’t really do differential compensation in any way. Regardless of what your broader theory of action is about the most promising ways to improve teacher quality, evaluations came to be seen as this prerequisite for them.

The reformers identified the passage of No Child Left Behind (NCLB), as a key policy window for this strategy. The law was formulated during the same time as researchers had begun developing robust econometric methods to assess the “value-added” of students being assigned to specific teachers—i.e., do a student’s learning outcomes, as indicated by their test scores, increase or decrease as a result of high or low-quality teacher performance? “Value-added” became a catchphrase for the kind of systematic teacher evaluation that the reformers sought—one based on quantitative data as opposed to solely observational data.

One interviewee, a thought leader, described this context: “It was driven by Ed Trust’s popularization of the Bill Sanders value-added stuff out of Tennessee, that was part and parcel of the birth of the NCLB.” This source refers to the work of William Sanders, who was an early author of studies on value-added models in education, as well as referencing the centrality of Education Trust, the respected education research and advocacy organization that was an outspoken advocate of value-added models and a grantee of major foundations. Another interviewee, a senior staff member of a prominent teacher quality advocacy organization, made a similar comment about the time period preceding NCLB: “Education Trust did a lot of work in that area when Gates was starting up.”

Simultaneously, foundations were pursuing a strategy that dovetailed with the passage of NCLB. Its accountability mandates for states to provide and utilize disaggregated data could not be executed without appropriate database systems. While many states had systems in place prior to NCLB, most did not have the capacity to process the massive influx of data that standardized testing would produce, including collecting, analyzing, reporting, and utilizing the data for improvement. Between 2000 and 2004, foundations made large grants to states and districts that were earmarked for developing computerized database infrastructure—a notable development given that foundations traditionally have not favored overhead costs such as investments in databases (Tompkins-Stange 2016). A 2004 report, targeted toward state policymakers, included the following summary of the political environment around data at the time:

Every state in the nation is taking a long, hard look at education data and how it can be collected, housed, analyzed, accessed, and used better than ever before…Most states have tried to meet these challenges head on, but, more often than not, demands have outstripped capacity as data needs continue to evolve and grow more complex. Policymakers will want to protect and improve their state’s information infrastructure in order to realize the full benefits of standards-based reform. (Palaich et al. 2004, p. 1)

Interviewees described this confluence of factors—state and foundation investments in databases, the subsequent availability of student-level data post-NCLB, and a concerted effort to promote test score-based evaluation—as a “perfect storm.” One interviewee, an education researcher, reflected on how these dynamics paved the way for the ascendance of value-added in policy debates: “Once you had NCLB data systems—which is amazing—you had the capability to calculate reading and math [scores] for five grades, [which] became the epicenter of Obama-style test-based teacher eval.” Concurrent investments in state-level big data provided the platform for value-added methods to spread, as this interviewee said: “We were talking about value-added as a much more sensible way to think about performance and proficiency levels. But the machinery wasn’t really in place.”

Roughly five years after the passage of NCLB, the Gates Foundation shifted away from small high schools, its flagship education program from 2000 to 2006, and Gates leadership redirected the foundation’s attention toward teacher quality and assessment. One interviewee, a policy analyst at an education advocacy organization, described the influence of Gates in the process of popularizing value-added methods: “[Our project] was largely funded by the Gates Foundation. They’d started this whole small schools [initiative], and then they looked for a switch and [entered into teacher quality].” Another interviewee, an education researcher at a think tank, stated: “The Gates folks decided that small schools weren’t the ticket. Then they went in ‘08 with Broad…and need[ed] something that looks and sounds different and value-added [satisfied that need].”

After the Obama Administration took office in 2009, a number of former Gates Foundation officials assumed senior roles in the Department of Education under Secretary Arne Duncan, and were influential in drafting Race to the Top, a $4.3 billion competitive grant program designed to induce states to comply with specific policy reforms, including the use of value-added methods in evaluation programs. The Department of Education’s call for proposals stated that Race to the Top grant winners would focus on advancing four specific reforms:

Adopting standards and assessments that prepare students to succeed in college and the workplace and to compete in the global economy; building data systems that measure student growth and success, and inform teachers and principals about how they can improve instruction; Recruiting, developing, rewarding, and retaining effective teachers and principals, especially where they are needed most; and turning around our lowest-achieving schools.

These implicit and explicit references to value-added measures and the need to evaluate and compensate teachers based on their effectiveness are evidence of the emergent debates around using student test scores to determine teacher pay—another plank of the education reformers’ theory of change. An interviewee from a foundation commented on the fact that after Race to the Top, states were required to “put together evaluation systems for teachers and states would begin to link this to hiring and firing.” The fact that this particular reform had acquired such political capital in a relatively short time was, in the words of this interviewee, “remarkable.”

Creating an evidence base

In addition to maintaining close networks with policy elites, foundations actively engaged in commissioning original research designed to provide an evidence base relevant to their policy priorities. Foundations make grants to intermediary organizations to conduct “advocacy research,” which has the explicit objective of being injected into policy discourse to be cited as empirical justification for desired reforms (Lubienski et al. 2009). Unlike traditional peer-reviewed research, which may pose uncertain conclusions regarding policy implications, advocacy research is shaped by specific policy objectives and political strategy and is typically produced by think tanks and nonprofit organizations, rather than universities (Shaker and Heilman 2004). The level of empirical rigor in advocacy research exists on a spectrum, from employing highly rigorous methods and considerations of external and internal validity, to omitting discussion of methods entirely.

While foundation-funded advocacy research is by no means the only source of policy-influential research in the teacher quality debate, it is central in Congressional hearings during our study period. Between 2000 and 2016, only nine research reports were cited three or more times by witnesses (and only one of which was peer-reviewed). The fourth-most cited report, which was consistently referenced in our interviews, was a 2009 advocacy research report by The New Teacher Project entitled The Widget Effect—a call to arms about the need for systematic teacher evaluation systems in order to distinguish between low-quality and high-quality teachers using test score-based evaluation methods. The report stated that “institutional indifference to variations in teacher performance” resulted in systems that perpetuated low-quality teaching across the country, taking aim at evaluation systems that relied predominantly on observational methods as opposed to econometric approaches (Weisberg et al. 2009). Several education reform-oriented foundations including the Gates Foundation, Walton Family Foundation, Robertson Foundation, and Joyce Foundation funded the report. Within a month of its release in 2009, Secretary Duncan made the following statement about the report in a speech:

These policies…have produced an industrial factory model of education that treats all teachers like interchangeable widgets. A recent report from the New Teacher Project found that almost all teachers are rated the same. Who in their right mind really believes that? We need to work together to change this.

The Widget Effect was praised by many interviewees as a triumph of advocacy research—a clear proposal and message, presented in a comprehensible and digestible format, that made a complicated issue more palatable. More importantly, however, the report was also a triumph for the policy networks surrounding teacher quality discourse—within a month, the report had had such impact that Secretary Duncan was referencing it in major speeches, which was accomplished by disseminating it through policy networks among actors with shared preferences.

The widespread recognition of The Widget Effect was emblematic of the rising prominence of advocacy research in policy debates. In the last ten years, education policy scholars have observed a shift toward targeted advocacy research funded by foundations, particularly surrounding issues of market-based policy interventions (Henig 2009; Lubienski et al. 2009). Contemporary examples of advocacy research contest the traditional conceptualization of expert researchers being separate and distinct from politics. According to Kingdon (2011, p. 228):

The policy community concentrates on matters like technical detail, cost-benefit analyses, gathering data, conducting studies, and honing proposals. The political people, by contrast, paint with a broad brush, are involved in many more issue areas than the policy people are, and concentrate on winning elections, promoting parties, and mobilizing support in the larger polity.

In current education policy networks, however, the converse is true, as researchers and advocates may overlap. One interviewee, a staff member of an education advocacy organization, described her role on a Gates Foundation-funded advocacy research project: “We saw a need to be more involved, not just from putting ideas out there but to help guide the conversation more hands-on.” Foundations, particularly those that endorse common education reform priorities, are now more likely to reject the traditional model of funding basic research in universities intended for diffusion into policy networks, but without the added leverage of a dedicated marketing structure to ensure, rather than impute, that the research reaches its intended audience.

This is particularly true for foundations that identify as strategic philanthropies  who are more likely to assertively use research as a tool to advance their policy goals. Strategic philanthropy is structured around the managerial concept of strategic planning, emphasizing clearly articulated goals from the outset of an initiative, the use of research to substantiate decisions, accountability from grantees in the form of benchmarks and deliverables as measured in incremental outcomes, and evaluation to assess progress toward milestones (Brest and Harvey 2008). Strategic funders also prioritize measurable returns on investments. Applying this formulation, basic research can appear very costly, with high levels of uncertainty or ambiguous returns, while targeted advocacy research promises better yield.

Interviewees described strategic foundations—most notably, the Gates and Broad Foundations—as highly influential leaders within the teacher quality policy network and depicted foundations’ theory of change as based on the assumption that teacher evaluation was necessary to advance other education reform goals, such as pay for performance and alternative teacher certifications. They also focused on these foundations’ use of research evidence as political in nature, departing from the “expert-led model of change” that Clemens and Lee (2010) describe and moving toward a model wherein researchers and advocates pursued similar goals: to inject policy ideas into political discourse more directly than their traditional philanthropic approaches.

Perhaps the most important example of this dynamic is the Gates Foundation’s investment in Measures of Effective Teaching (MET), a longitudinal study that sought to test and validate methods to systematically identify high-quality teaching. MET used a mix of methods, including value-added, classroom observations, and video and audio recording of nearly 3000 teachers. While the MET has been lauded for its unprecedented scale and scope, some scholars have also critiqued its validity, arguing that the study was, at least to an extent, part of a broader strategy to generate evidence in support of value-added methods, in order to provide a research base for advocacy on teacher quality at the state and federal levels. One interviewee, an academic researcher, summarized the critique of the MET:

It’s presented in a way that’s pushing for conclusions stronger than the research can support…There was a demand for [test score-based teacher evaluation policy] and then people went and found research that supported it.

This view, while expressed by a number of interviewees, was interpreted differently among scholars. One source acknowledges the “backward” process of generating evidence to support a predefined outcome, but describes it as a type of public service that the Gates Foundation provided:

It was a noble effort to policymakers who were headed in a given direction already through the adoption of these statewide systems. The Obama Administration was committed to that and this was an attempt to give them some information to work with.

Another researcher made the same argument, but framed it more cynically:

It was pretty clear [during] the MET study that [the researchers] needed to get the result that the Obama administration wanted. He was doing this because of outside stakeholders wanting some answer, wanting to know how to weigh these evaluations, wanting some evidence about what should be the most important or wanting evidence on validity.

Foundations’ growing preference for advocacy research was apparent in our interviewees’ accounts of their interactions with the Gates Foundation as well as other funders, who sought to distill the complicated technicalities of test score-based teacher evaluation into a format that policy elites could easily understand. This framing positioned value-added as commonsensical, as one interviewee, a think tank leader, stated: “Early on I remember even Randi [Weingarten, president of the American Federation of Teachers] said, ‘Well, that seems to make sense.’” Multiple interviewees echoed this insight. For example, the leader of an education consultancy described the appeal that value-added had as a technical solution during the late 2000s:

[Value added] was very central to the discourse. It was done, actually, as central to the policies. It became all you heard about. It became the thing everybody talked about…the [policy elite] didn’t understand [how to address the issue of low-quality teachers], and they thought [value-added] was going to be a panacea.

Interviewees described how, consistent with quantitative findings about the shared policy preferences of grantees, philanthropists were key to elevating the idea of value-added as a central policy preference. One interviewee, a leader of a research organization, described value-added as a concept that “Bill Gates got fascinated” with “because it was something a technocrat can get excited about.” Another source, a former Gates Foundation staff member, shared the following anecdote:

You know how legends get told about these things. There’s a legend–I think this is actually really true–where Bill Gates met with [several proponents of value-added evaluation] in [a large city] and basically showed him, Bill, these numbers, and he was all in. And that’s why he ended up getting three hundred million dollars invested in this strategy and another forty million in the research.

These statements align with prior findings by Tompkins-Stange (2016), who quoted a Gates Foundation official discussing how research reports used to advocate for value-added sometimes lacked rigor, or were based on assumptions that did not apply to all districts or states:

As people get away from technical appendices and the caveats… and distill those into summaries and into Power Points, into bullet points, and then into data to inform decisions, all that gets washed away and comes back down to…what was originally the presupposition you began with which is “Value-added is the way to evaluate and pay teachers.”

Thus, value-added teacher evaluation resonated with key policy entrepreneurs in the teacher quality discourse network, facilitating its rapid and central emergence in policy debate. This prominence, however, also had the effect of dominating debate and engendering a backlash. A former Gates Foundation official described how originally, teacher evaluation was only one of five education “levers,” or pathways to desired change, that the education team identified; others included directing more resources to teacher leadership and instructional innovation. Of these five levers, however, teacher evaluation received the lion’s share of attention, both inside the foundation and among the broader public and press. This interviewee described:

So the message became all about the teacher human capital. I think we lost the messaging around children, teachers, important relationships, and great content. And the way we came out was research and accountability, and the Eric Hanushek stuff.

This comment about Eric Hanushek, a senior education economist and author of seminal studies on value-added, refers to a statement Hanushek has made regarding the use of value-added as a tool to identify the lowest performing 5–8% of teachers and replace them. Hanushek (2011) frames this objective in economic terms, arguing that eliminating low-performing teachers would improve future individual earnings and economic growth:

A teacher one standard deviation above the mean effectiveness annually generates marginal gains of over $400,000 in present value of student future earnings with a class size of 20 and proportionately higher with larger class sizes. Replacing the bottom 5–8% of teachers with average teachers could move the U.S. near the top of international math and science rankings with a present value of $100 trillion.

While this message resonated with policymakers and funders, teachers and teachers’ unions interpreted it as an indictment of their profession. In perhaps the most high-profile example of this controversy, a group of Houston teachers sued the district in federal court in 2011 (Houston Federation of Teachers vs. Houston Independent School District) arguing that their value-added scores were being used inappropriately, and provided the central justification for the firing of 221 teachers in a single year. In 2017, the lawsuit settled, with the district agreeing to end the use of value-added for high-stakes teacher evaluation.

Several interviewees from the Gates Foundation expressed frustration that the field had coalesced around the message of “the 5–8%” of poorly performing teachers that Hanushek and others advocated for replacing, rather than taking a more holistic approach. Their interpretations of the backlash were substantially different. One former staff member viewed the backlash as an intentional effort by teachers unions to stymie evaluation efforts, arguing that in practice, the use of value-added to inform high-stakes employment decisions was not political, but rather technical:

The notion of trying to have evaluation “count for something”—a person’s ability to maintain their job—ended up in the courts. That really ignited a lot of consternation and anger. The fact that your ability to be laid off is a time stamp date on a certain hour of the day you get employed, is no different than we just say “Let’s get rid of the shortest teachers” when we go to do the layoffs. I think that added another box to the conversation intensity about all this.

Another official had a contrasting take on how the controversy unfolded, reflecting on the broader cultural antecedents that informed teachers’ resistance. One former Gates Foundation official shared:

I think the field has changed dramatically… no one’s fighting about [teacher evaluation feedback systems] anymore. There’s a shared understanding of what we mean by “effectiveness” in this system. That didn’t exist years ago [and] is a huge accomplishment. But, in the process, there was collateral damage—the [teaching] profession took more of a hit. There was too much teacher blame. I think we got ahead of it after a while, but this was a baby that was born under a cloud. And so, as the child grew up, it was hard to pull the child from under the cloud.


In this study, we utilized a mixed method approach to understand whether foundations assumed roles as policy entrepreneurs in the realm of teacher quality policy. Through innovative new methods, we quantified the extent to which foundation funding was associated with the development of shared policy preferences on the part of grantees. Through qualitative interviews, we explored how funders’ involvement played out on the ground, from the perspectives of key actors in the teacher quality policy network.

Through the sponsorship of advocacy research, foundations advanced their policy preferences in two ways:
  • Using big data and the availability of new evaluation systems in states, funders developed and applied systematic indicators and evaluative frameworks to generate evidence that could bolster or validate key policy preferences.

  • Galvanizing political support for policy preferences by sponsoring advocacy research to diffuse evidence about the effectiveness of these approaches.

These mechanisms were the result of significant investments foundations made to construct a policy network and positioning themselves as entrepreneurs within the network who could commission research evidence and fund organizations to inject it directly into the policy debate, ensuring it reached both policy elites and the broader advocacy community. Through this systematic approach, foundations assumed a key role in placing policy preferences for value-added teacher evaluation on the federal political agenda. Our quantitative findings demonstrate that foundation grantee witnesses in Congressional hearings were significantly more likely to share policy preferences related to teacher evaluation than their non-grantee counterparts. Thus, foundation grantee witnesses differed systematically in their policy preferences compared to non-grantees.

This study provides a touchpoint for further study to establish the directionality of the causal arrow between foundation sponsorship and the behavior of grantees. One of our interviewees described, it is nearly impossible to credibly assess whether foundation funding or endorsement drives field behavior or merely channels it. While our results point to a plausible interpretation of foundations as policy entrepreneurs that are uniquely positioned within policy networks, this entrepreneurship could have unfolded through multiple mechanisms. Prior research has indicated that foundations drive change from the “top-down,” pursuing elite, expert-driven agendas and enrolling grantees to execute foundation-determined strategies (Reckhow 2013; Tompkins-Stange 2016). However, our interviews suggested that foundations were also sponsoring key advocacy organizations with preexisting policy preferences that aligned with their philanthropic agendas, subsequently amplifying their influence and creating potential ripple effects for other potential grantees, but not dictating their behavior. Thus, foundations are not necessarily the independent architects of a new policy consensus, but rather the facilitator of or catalyst for an already-emerging policy consensus.

These limitations notwithstanding, this mixed method study breaks new ground in providing quantitative evidence about the existence of aligned policy preferences among philanthropically funded groups, and qualitative evidence that reveals how and why these preferences originated, traveled in policy networks, and were adopted formally into policy. Taken together, the findings from both quantitative and qualitative evidence support the argument that foundations were highly central actors within teacher quality policy, that the Gates Foundation was especially influential in this policy area, and that funder-supported priorities rose to the top of policy discourse. The qualitative data further highlights how the Gates Foundation deliberately sought to use research as a form of political advocacy, along with its position as policy entrepreneur within the teacher quality policy network, to do so. Free from the constraints of government, rich in resources and prestige, and closely networked with policy elites, foundations were able to legitimate and disseminate the idea of value-added teacher evaluation—an idea that had political promise due to its resonance with policy elites’ managerial expertise, and within the context of a cultural push for evidence-based policy.


  1. 1.

    Our decision to exclude the question and answer segments of the hearings is based on Fisher, Waggle, and Leifeld’s approach to discourse network analysis of Congressional testimony on climate change. We elected to analyze only witness statements that demonstrate specific policy preferences, as opposed to the back-and-forth dialogue between witnesses and policymakers. Like Fisher, Waggle, and Leifeld, we determined that this decision would provide a reliable and systematic process for coding. Witness statements are entered into the record with consistency in duration, form and style, whereas question and answer portions are frequently dominated by some voices over others; are dependent on background context that is not included in the record, nor that can be specifically linked to a single witness; and vague or off-topic. This methodological decision has implications for our analysis; had we elected to include the question and answer sections for all hearings, we would no doubt capture additional nuance that emerges in debate and dialogue, which might deepen and clarify our inferences. The difference between these two research designs poses a ripe area of future research.

  2. 2.

    The foundations included are: Gates, Broad, Walton, Arnold, Joyce, Carnegie, Robertson, Kellogg, Dell, Silicon Valley Community Foundation, Hewlett, GE, Irvine, Fisher, Communities Foundation of Texas, Daniels, and Ford.



Funding was provided by William T. Grant Foundation (Grant No. 183183). The authors would like to thank Sarah Galey for her important contributions to the project and research assistance. We would also like to thank Abigail Orrick and Jeffrey Snyder for their research assistance.


  1. Ansell, Chris, Sarah Reckhow, and Andrew Kelly. 2009. How to reform a reform coalition: Outreach, agenda expansion, and brokerage in urban school reform. Policy Studies Journal 37(4): 717–743.CrossRefGoogle Scholar
  2. Baumgartner, Frank, and Bryan Jones. 2012. From There To Here: Punctuated Equilibrium to the General Punctuation Thesis to a Theory of Government Information Processing. Policy Studies Journal 40(1): 1–20.CrossRefGoogle Scholar
  3. Berry, Jeffrey. 2002. Validity and Reliability Issues in Elite Interviewing. Political Science and Politics 35(4): 679–682.CrossRefGoogle Scholar
  4. Borgatti, S.P., M.G. Everett, and L.C. Freeman. 2002. UCINET for Windows: Software for Social Network Analysis. Harvard, MA: Analytic Technologies.Google Scholar
  5. Brest, Paul, and Hal Harvey. 2008. Money Well Spent: A Strategic Plan for Smart Philanthropy. New York, NY: Bloomberg Press.Google Scholar
  6. Bushouse, Brenda. 2009. Universal Preschool: Policy Change, Stability, and the Pew Charitable Trusts. Albany, NY: SUNY Press.Google Scholar
  7. Clemens, Elisabeth S., and Linda C. Lee. 2010. Catalysts for Change: Foundations and School Reform 1950–2005. In American Foundations: Roles and Contributions, ed. Helmut Anheier and David C. Hammack. Washington, DC: Brookings Institute Press.Google Scholar
  8. Dexter, L. 1970. Elite and Specialized Interviewing. Evanston, IL: Northwestern University Press.Google Scholar
  9. Duncan, A. 2009. Partners in Reform. Remarks of Arne Duncan to the National Education Association. Accessed from Accessed 28 Aug 2018.
  10. Fisher, Dana R., Joseph Waggle, and Philip Leifeld. 2013. Where Does Political Polarization Come From? Locating Polarization Within the U.S. Climate Change Debate. American Behavioral Scientist 57(1): 70–92.CrossRefGoogle Scholar
  11. Galey, S., Reckhow, S., and Ferrare, J. 2017. Idea Brokerage in Teacher Policy Networks. Paper Presented at the Annual Meeting of the American Educational Research Association.Google Scholar
  12. Goldstein, K. 2002. Getting In the Door: Sampling and Completing Elite Interviews. Political Science and Politics 35(4): 669–672.CrossRefGoogle Scholar
  13. Goldstein, D. 2014. The Teacher Wars: A History of America’s Most Embattled Profession. New York, NY: Knopf Doubleday.Google Scholar
  14. Hanneman, R.A., and M. Riddle. 2005. Intoduction to Social Network Methods. Riverside, CA: University of California, Riverside.
  15. Hanushek, Erik. 2011. The Economic Value of Higher Teacher Quality. Economics of Education Review 30(3): 466–479.CrossRefGoogle Scholar
  16. Henig, Jeffrey R. 2009. Politicization of Evidence: Lessons for an Informed Democracy. Educational Policy 23(1): 137–160.CrossRefGoogle Scholar
  17. Hwang, H. and W. Powell. 2005. Institutions and Entrepreneurship. In Handbook of Entrepreneurship Research, ed. S. A. Alvarez, R. Agarwal and O. Sorenson, 23. New York, NY: Kluwer Publishers.Google Scholar
  18. Jenkins-Smith, Hank C., Daniel Nohrstedt, Christopher M. Weible, and Paul A. Sabatier. 2014. The Advocacy Coalition Framework: Foundations, Evolution, and Ongoing Research. In Theories of the Policy Process. 3rd ed. Paul A. Sabatier and Christopher M. Weible. Boulder, CO: Westview Press.Google Scholar
  19. Jones, Bryan D., and Frank R. Baumgartner. 2005. The Politics of Attention: How Government Prioritizes Problems. Chicago, IL: University of Chicago Press.Google Scholar
  20. Kingdon, John W. 2011. Agendas, Alternatives, and Public Policies. London, UK: Longman Publishing Group.Google Scholar
  21. Kirst, Michael W., and Frederick M. Wirt. 2009. Political Dynamics of American Education, 4th ed. Richmond: McCutchan Publishing Corporation.CrossRefGoogle Scholar
  22. Leifeld, Philip. 2013. Reconceptualizing Major Policy Change in the Advocacy Coalition Framework: A Discourse Network Analysis of German Pension Politics. Policy Studies Journal. 41(1): 169–198.CrossRefGoogle Scholar
  23. Lubienski, Christopher, Peter Weitzel, and Sarah Theule Lubienski. 2009. Is there a ‘Consensus’ on School Choice and Achievement? Advocacy Research and the Emerging Political Economy of Knowledge Production. Educational Policy. 23(1): 161–193.CrossRefGoogle Scholar
  24. Meier, Kenneth, and Laurence O’Toole. 2006. Bureaucracy in a Democratic State: A Governance Perspective. Baltimore, MD: Johns Hopkins University Press.Google Scholar
  25. Merriam, S. 2009. Qualitative Research: A Guide to Design and Implementation. San Francisco: Jossey-Bass.Google Scholar
  26. Moe, Terry. 2011. Special Interest: Teachers Unions and America’s Public Schools. Washington, DC: Brookings Institution Press.Google Scholar
  27. Palaich, Robert, Dixie Griffin Good, and Arie van der Ploeg. 2004. “State Education Data Systems That Increase Learning and Improve Accountability. Policy Issues: Research Based Analysis of Education Policy 16: 1–11.Google Scholar
  28. Quinn, Rand, Megan Tompkins-Stange, and Debra Meyerson. 2014. Beyond Grantmaking: Philanthropic Foundations as Institutional Entrepreneurs and Agents of Change. Nonprofit and Voluntary Sector Quarterly 43(6): 950–968.CrossRefGoogle Scholar
  29. Reckhow, Sarah. 2013. Follow the Money: How Foundation Dollars Change Public School Politics. New York, NY: Oxford University Press.Google Scholar
  30. Reckhow, Sarah, and Megan Tompkins-Stange. 2015. “Singing from the Same Hymnbook” at Gates and Broad. In The New Education Philanthropy, ed. Frederick M. Hess and Jeffrey R. Henig. Cambridge, MA: Harvard Education Press.Google Scholar
  31. Reckhow, Sarah and Jeffrey Snyder. 2013. The Education Philanthropy Network: Convergence Toward One Dominant Model. Paper presented at the American Educational Research Association Annual Meeting. San Francisco, CAGoogle Scholar
  32. Rhodes, R.A.W. 2006. Policy network analysis. In The Oxford handbook of public policy, ed. M. Moran, M. Rein, and R.E. Goodin, 423–445. Oxford, UK: Oxford University Press.Google Scholar
  33. Scott, Janelle and Huriya Jabbar. 2014. The Hub and the Spokes: Foundations, Intermediary Organizations, Incentivist Reforms, and the Politics of Research Evidence. Educational Policy, 28(2): 233–257.CrossRefGoogle Scholar
  34. Shaker, P., and E. Heilman. 2004. The new common sense of education: Advocacy research vs. academic authority. Teachers College Record, 106(7): 1444–1470.CrossRefGoogle Scholar
  35. Tompkins-Stange, Megan. 2016. Policy Patrons: Philanthropy, Education Reform, and the Politics of Influence. Cambridge, MA: Harvard Education Press.Google Scholar
  36. Weible, Christopher, and Paul Sabatier. 2014. Theories of the Policy Process, 3rd ed. Boulder, CO: Westview Press.Google Scholar
  37. Weisberg, Daniel, Susan Sexton, Jennifer Mulhern and David Keeling. The Widget Effect: Our National Failure to Acknowledge and Act on Differences in Teacher Effectiveness. The New Teacher Project, 2009.Google Scholar
  38. Wilson, James Q. 1991. Bureaucracy: What Government Agencies Do and Why They Do It. New York, NY: Basic Books.Google Scholar
  39. Wood, B.D., and N.A. Theobald. 2003. Political responsiveness and equity in public education finances. Journal of Politics 65(3): 718–738.CrossRefGoogle Scholar

Copyright information

© Springer Nature Limited 2018

Authors and Affiliations

  1. 1.Michigan State UniversityEast LansingUSA
  2. 2.University of MichiganAnn ArborUSA

Personalised recommendations