A Comparative Analysis of How Actors Implement: Testing the Contextual Interaction Theory in 48 Cases of Wetland Restoration

Abstract

This paper applies a two-actor, causal, deductive theory of implementation (the contextual interaction theory) to analyze 48 wetland restoration cases. The theory uses motivation, information, and power of the policy implementer and target to predict the nature of the implementation process (e.g. cooperation, obstruction, etc.). The research question centers on the predictability potential of the contextual interaction theory. It asks if the theory accurately predicts process interactions, based on the quantification of actor motivation, information, and power. In the analysis, a strong correlation was found between expected and observed results, or a high predictability potential. To overcome limitations, a formula version was also tested using a correlation design. These results did not produce an equivalent fit. One explanation is that implementation entails threshold values in the core variables, and a broad neutral category that fails to influence action in the same manner. This theory is found to be a useful tool for consistent, comparative, and replicable analyses that should be applied more broadly in implementation studies to better understand its strengths and limitations.

Introduction

When filtering public policy through the circumstances of reality, intention may not correspond with outputs. Empirical research discloses implementation as a significant impediment to change (Pressman and Wildavsky 1973; Palumbo and Calista 1990; Hill and Hupe 2002; Bressers 2004). This paper contributes to implementation studies by testing a candidate analysis tool, the contextual interaction theory. The theory evaluates how actor characteristics influence the implementation process by analyzing 48 wetland restoration cases in the European Union and the United States.

Wetlands are the empirical focus of this project due to their importance in improving water quality, enhancing flood control, and providing wildlife habitat (Tiner 1998; Lewis 2001; Schuyt and Brander 2004). In both Europe and the United States, as population increased over time and as people shifted to stationary agricultural practices, wetlands were filled or transformed to make them “more useful” (OECD 1996; Vileisis 1997). More recently, policy makers, governments, organizations, and citizens have understood that the world's wetlands require some level of restoration to protect their important functions. Policies to restore wetlands can be found internationally and at all levels of government. Refining the implementation of these policies and programs can enable more efficient and effective restoration.

Wetland restoration projects integrate a broad range of policies, while justification for restoration may connect with one or a number of issues (e.g. safety, wildlife, habitat, recreation). It is not safe to assume when dealing with wetlands that a governmental organization will implement a policy. For this reason it is important to have an open approach when considering who plays the roles of implementer and target. For all of these reasons, wetlands are a complex empirical field, and as such represent a strong test for the theory in application.

Defining Implementation

To understand implementation it is important to define the concept as used in the scope of this project. In their early work on implementation, Pressman and Wildavsky (1973) begin with an implementation definition taken from Webster's 1969 Dictionary: “to carry out, accomplish, fulfill, produce, complete” in this context, “a policy” (xiii). Implementation researchers continue to create and use definitions of implementation more specific to its policy application, emphasizing a number of concepts. Mazmanian and Sabatier (1983) define implementation as “the carrying out of a basic policy decision” (20). Wittrock (1985) finds that implementation means “translating policy commitments and societal aspirations into real world effects” (17). Palumbo and Calista (1990) describe implementation as “a series of interactions and interpretations between the outputs of policy formation and the effects of organizational and inter-organizational impacts, between the latter and street-level bureaucratic behaviors, and between the latter and target group behaviors” (xiv). O'Toole (2000) defines implementation as “what develops between the establishment of an apparent intention on the part of government to do something, or to stop doing something, and the ultimate impact in the world of action” (266). For the purpose of this study we define implementation as translating public policy intention into results.

Characterizing Implementation

Many scholars find it fruitful to treat implementation as a distinctive point for analysis within the policy process with an ability to shed light on the whole (Pressman and Wildavsky 1973; Palumbo and Calista 1990; Hill and Hupe 2002; Bressers 2004). In their 1973 study, Pressman and Wildavsky found a lack of connection between policy goals and results when examining one federal policy in California. In this vein, the first generation of implementation research consisted chiefly of solitary case studies featuring negative reports of the way governments implement their own programs (Sabatier 1986). The second generation featured an increase in comparative analysis, seeking to clarify differences in policy implementation through a focus on precise variables and theoretical frameworks, generally upholding the top down perspective of first-generation work (Sabatier 1986). Critics found this body of research failed in developing testable, elucidatory theory and in creating a thorough, fused methodology (Schofield 2001). Goggin et al. (1990) promote a third generation of research to illuminate the variability within implementation scenarios by using more stringent scientific methods.

In the past, theories approached implementation from either the top down or bottom up perspectives. The top down approach asserts that once policy objectives are set forth in legislation the implementation process follows linearly (Mazmanian and Sabatier 1983; Schofield 2001). In contrast, the bottom up approach highlights local implementers who tend to emphasize the problem instead of strict policy goals (Hjern 1982; Schofield 2001). After years of debate ¹ between proponents of top down and bottom up approaches, most researchers concede the merits of both perspectives, with several scholars recommending synthesis of the two into a unifying model (Elmore 1982, 1985; Wittrock 1985; Sabatier 1986; Bressers et al. 1989; Goggin et al. 1990; Bressers and Ringeling 1995; Matland 1995; O'Toole 2002). One can briefly characterize the field of implementation since its inception as producing a great deal of research, including many case studies featuring a range of variables and having a heavy emphasis on inductive approaches. This is not to say that large-N empirical work is nonexistent but that it has failed to produce a parsimonious, generally accepted explanatory theory of implementation interactions (O'Toole 2000). This large-N study was created in answer to Goggin et al.'s quest for a third generation of implementation research.

The Contextual Interaction Theory

Justification of CIT

The contextual interaction theory (CIT) was developed as a theory of implementation in the Netherlands during the late 1990s and has been applied in several studies since that time but is not widely used (Dinica and Bressers 2003; de la Torre 2006; van Tilburg 2007; Owens 2008; Spratt 2008). Relatively simple, with broad applicability, the theory analyzes the very core of implementation: the motivation, information, and power of target and implementer. CIT emphasizes the policy target and implementer, whether they exist as local implementers or high-level administrators. CIT is an example of a “third generation theory” (Goggin et al. 1990) with capacity to bridge the top down and bottom up divide by centering the analysis on the interaction process of implementer and target, whether a government administrator, or an on-the-ground stakeholder. At the same time it incorporates an external criterion of success (not just participant satisfaction, as in many bottom-up approaches) but societal “goods” that may also, but need not be, the official goals of policy.

A useful theory must condense reality into less detailed but informative elements. Scharpf (1997) writes that overly parsimonious theories “ignore” either actors or institutions in pursuit of the other (36). An all-inclusive approach may present a realistic depiction that lacks descriptive simplicity and comparability. CIT is parsimonious, distilling a sea of options for implementation variables into three core variables of motivation, information, and power. It is important to note that these variables are not arbitrarily chosen as three important variables among others, but because they have high explanatory power and exist at the core of interaction processes (Bressers 2004; Owens 2008). Motivation (Gross et al. 1971; Ball 1976; Larson 1980; Nakamura and Smallwood 1980), information (Baum 1976; Edwards 1980; Williams 1980; Baum 1981; Williams 1982), and power (Bunker 1972; Berman 1980; Raelin 1980, 1982; Ackermann and Steinmann 1982; Davies and Mason 1982; Browne and Wildavsky 1984) have been identified as key variables for implementation by other scholars. All three consistently emerge as core ideas within O'Toole's (1986) list of hundreds of variables deemed “important” by researchers in the field (Owens 2008). We argue that the theory's parsimony is rooted in the explanatory power of the variables and that the theory provides parsimony without a great expense in realism. CIT is therefore a deductive and realistic approach that allows implementation to be effectively analyzed.

This study's objective is to understand how successfully the theory predicts interaction processes. We test the theory by applying it to an empirical field dissimilar to its field of origin (i.e. in permitting and subsidizing). We also expand the theory by applying it, a two-actor model, in multi-actor cases. To do so we assume that within each case actors can be divided along meaningful lines into coalitions, with the understanding that these can be formal or informal groups. These actors can be identified based on core preferences within each case. As described by Sabatier and Weible (2007) actors often divide over policy core policy preferences, or beliefs that “project an image of how the policy subsystem ought to be, provide the vision that guides coalition strategic behavior, and helps unite allies and divide opponents” (195). In other words, we test the viability of a two-actor simplification in a policy area that does not always lend itself to a strict two-actor scenario to better understand if this model is capable of elucidating a multi-actor scenario.

Description of CIT

The theory focuses on motivation, information, and power of policy implementers and targets. The implementer is the actor “officially commissioned with promoting the envisaged measures”, and the target is the “actor necessary to realize [the measures]” (Bressers 2005: 1). Based on various combinations of the independent variables (motivation, information, and power), the theory produces a prediction of the type of process interaction that will occur (e.g. passive cooperation, forced cooperation, obstruction). The 14 potential interactions can be seen in the flowchart found in Figure 1, with corresponding predictions available in Appendix 1. In this study we ask whether CIT accurately predicts process interactions based on motivation, information, and power of the target and implementer. CIT produces the predictions in Appendix 1 for each combination of the independent variables, as shown step-by-step in the flowchart. The results do not conform to predictions in the strictest sense, since they are developed ex post.

Figure 1. Contextual interaction theory flowchart

Independent Variables ²

Conceptualization of the variable motivation incorporates themes such as an actor's own motivation as well as potential sources of external pressure. Own motivation includes aspects such as compatibility with the goals of implementation, work-related motivation, the actor's attitude to the implementation objective, attitude to the target group, and a question about self-effectiveness. (During this process, if something is important to your group and others disagree, what do you think are your chances of attaining goals important to you?) Understanding potential sources of external pressure includes examining normative, economic, social, and political influences.

Conceptualization of information includes general knowledge about the policy and how to comply, accessibility to informational materials, and the transparency of the process for both targets and implementers. Questions consider both reported lack of and possession of information. General knowledge includes aspects such as policy awareness for relevant actors, including an understanding of policy requirements and benefits, and knowledge of other stakeholders and their role in the process. Transparency incorporates accessibility and the level of documentation available to process participants or interested parties. It also encompasses the simplicity or usefulness of this information. Creation of this variable also considers a dearth of transparency or accessibility. In other words, this includes having difficulty in finding or using information or encountering uncertainties that influence the process. In addition, questions are asked about information the actor would like to have had, to understand information missing from the process.

The conceptualization of power in this application of the theory includes both capacity and control. An organization's power is not static, and therefore it is important to understand its capacity, particularly in terms of resources such as finances, personnel, or time support. Resources (or a lack of them) have the ability to strengthen (or weaken) the position of a given actor. Power as control harkens to the more common understanding of the word – an ability to control, or force, a less powerful actor to act. Power as control is made up of formal and informal facets. Formal power is that given to a group, individual, or agency through legal channels or areas of responsibility. Informal power, which may counteract formal sources, can derive from roles (i.e. as site users or stakeholders) or from the ability to use expertise, coalitions, or media to one's advantage. Informal sources of power may also stem from the ability to convince others to comply with one's goals. When considering formal or informal power, it is important to reflect on the difference between power and reputation of power. Reputation of power incorporates how actors perceive each other in the process. In essence, the reputation of power is real to actors unless or until experiences prove that the reputation is not grounded in reality. For this reason, it is extremely important to understand how the actors comprehend their own power in relation to that of others in the process.

Dependent Variables

The dependent variable is the type of process interaction that is predicted to occur based on varying combinations of the independent variables, and can take one of many forms (descriptions from Bressers 2004). Active cooperation occurs when both parties share a common goal (remembering that the goal can also involve an attempt to halt the project). Passive cooperation happens when one of the parties adopts a relatively passive stance which neither hinders nor stimulates the project. Forced cooperation is a form of passive cooperation that is imposed by a dominant actor. Opposition occurs when one of the actors tries to prevent application by another actor. Obstruction occurs when one of the actors prevents application by another actor. Learning towards… occurs when only a lack of knowledge stands in the way of application. No interaction: There are also situations in which there will be no interaction at all between the implementer and the target groups.

Methodology ³

The unit of analysis is a given wetland project. The policy/project goal in each case is wetland restoration. The sample includes 12 cases from each of the focus areas (The Netherlands, Finland, New Jersey, and Oregon). We built a potential case list for each state by pinpointing agencies and programs promoting restoration. In the US, cases were found through the EPA's River Corridor and Wetland Restoration Project Directory (United States Environmental Protection Agency 2007a), which included cases from 1985 to 2003, and the 5 Star Grant Program Directory (United States Environmental Protection Agency 2007b), which included New Jersey cases from 1999 to 2005 and Oregon cases from 1998 to 2005. For the European Union cases, projects were found by way of the European Union LIFE projects database (Europa 2007) including all wetland restoration proposals granted in Finland or the Netherlands from 1992 to 2004.

Using these initial case lists and a snowball sampling approach, we created a list of 48 cases, 12 from each state, including three key types of cases within each state. The cases for inclusion fell into three broad categories: implemented (four per state), not yet implemented (four per state), and not implemented (four per state). This was not a proportionate stratified sample, but a device to ensure incorporation of both successful and unsuccessful cases. Through snowball sampling the initial lists grew and changed to eventually produce 12 cases per state fitting the study parameters.

We chose this 4 × 12 design to be able to also allow comparison of two jurisdictions (liberal US states and social EU states) and two population densities (Oregon and Finland (low density) versus the Netherlands and New Jersey (high density)). We asked whether CIT would do equally well in all four circumstances and whether the division of various process types would differ.

Data collection included semi-standardized interviews, or using a set of predetermined questions asked in a systematic manner with the expectation that the interviewer probe for deeper descriptions (Berg 2001). It became clear that for analysis via contextual interaction theory, obtaining surveys for both the project implementer and target would be essential. Realizing the typical low response rates and the limited pool of available cases, this method seemed inappropriate. At the same time, it was important to find an efficient and timely way of obtaining the data necessary for analysis.

To identify interviewees first we communicated with the person listed as the primary contact for each case. After initiating communication with several “primary contacts” it was apparent that the individuals listed on project descriptions often knew little in practice about the projects. For some projects, these individuals existed at the top of the funding chain. While they could not always speak personally about the case, they could refer to someone knowledgeable about the project. Occasionally there were several rounds of referral until someone directly knowledgeable about the case was reached. In many instances, this was a person within the agency or a scientist assigned to the project.

In practice we found that the individual knowledgeable about the case fulfilled the role of a core actor (either target or implementer). At the end of each interview, we used a snowball method to find other actors intimately associated with the case. One of these contacts yielded the other core actor. We conducted 96 interviews for this study. In 46 cases, two interviews were completed for each case (yielding 92 interviews), for two Finnish cases only one interview was possible (therefore analysis was impossible), and in another two cases a third interview occurred. Therefore, of the 96 interviews sought (two for each of 48 cases), 94 were completed, providing a 98 per cent response rate. As compared to traditional survey methods, this approach provided not only the information most suited for analysis (i.e. implementer and target perspectives) but also yielded an extremely high response rate.

The 96 interviewees included politicians, ranchers, land owners, representatives of government at all levels, non-profits, and farmers' unions, and other actor types. The majority were via telephone, ranging from 45 minutes to two hours. Of the 96 interviews, nine were submitted by email, an option offered to and used only by Finnish actors reluctant to submit to a telephone interview given their comfort level with English. Any omissions in response from emailed interviews were clarified with follow-up emails or telephone calls. Seven interviews were held in-person.

A conscientious researcher remains aware of the potential for actors to present past events and their own roles in the most flattering light. The interview instrument features several questions meant to triangulate responses or ask for details about stated actions, with the intention of creating checks and balances as actors present their stories. We treated each actor interview with a critical eye, insuring all data was fully supported by the actor's own responses and other accounts of events.

Analysis

Motivation, information, and power scores are based on calculating proportional responses to interview questions. ⁴ The interview instrument includes 26 points of assessment for motivation, 17 for information, and 22 for power. We evaluated the variable scores using the CIT flow chart, which then provided a prediction of the interaction that ensued. In measuring the dependent variable, in each case we created a case summary, case history, as well as an analysis of the roles of actors in the process. We compared the prediction to the case history and summary, to evaluate whether case results agreed with the CIT prediction.

In our first analysis, we explore the predictability potential of the theory by testing the relationship between observed and predicted results. Our second analysis uses statistical analysis to highlight relationships between variables within the sample using a quantitative/formula-based approach. ⁵ Finally, we analyze subsets of the cases to understand the influence of state type and population density on predictability potential.

Analysis One: Results

First we test the predictability potential of the theory by comparing observed and predicted results. Within the sample there were 27 cases of active cooperation (56 per cent), eight cases of opposition (17 per cent), three passive cooperation cases (6 per cent), three obstruction cases, two cases of learning (4 four), two cases with no interaction, one case of forced cooperation (2 per cent), and two incomplete cases. When looking at these cases in total, for well over half (64 per cent) we observe active cooperation, passive cooperation, or forced cooperation, and in far fewer (23 per cent) we observe non-cooperative interactions such as opposition and obstruction. In addition, there are a few cases of no interaction, or learning toward another interaction (4 per cent). In only one case was the theoretical expectation not in accordance with the observed reality of the project (Farnham Park). Therefore in an overwhelming proportion of the cases CIT generated predicted interactions that described the realities of the case process well. This overview also clearly shows that each state produces a variety of interaction types. In the most basic sense, theoretical expectations match observations in 45 of the 46 (98 per cent) of cases.

We asked whether CIT would do equally well in all four states, which it did. We also questioned whether high population density states would have more conflict, and low population density states less, but we found the four states did not show any significant variation in this regard. In terms of cooperation, the states had similar percentages of cooperative cases New Jersey (66.7 per cent), The Netherlands (58.3 per cent), Oregon (66.7 per cent), and Finland (70 per cent).

Analysis One: Discussion

It is important to note that the sample is most likely skewed toward the over-inclusion of cooperation cases, as all “implemented” cases register as active cooperation or passive cooperation. In addition “not yet implemented” cases may also be unrealistically skewing toward cooperation. In several cases, interviews take place early in the process, and earn the designation of a form of cooperation. In reality analysis may be occurring before oppositional actors become involved. In this analysis, comparing expected and observed interactions indicates a high percentage of cases where the expected results match reality. This is strong evidence of the usefulness of the CIT as an ex post explanatory tool.

Analysis Two: Results

In our second analysis we substitute a formulaic representation of actor scores to understand if this provides a better picture of reality. When using the flowchart model, motivation scores are listed as positive, neutral, or negative, the information level of the most positive partner is assessed, and the balance of power between actors is considered. In other words, the flowchart takes quantified information, then categorizes and compartmentalizes it. While this may make the elements of a given score less distinct, ⁶ the potential gain of having continuous values for explanatory factors and outcomes make it worthwhile to test. It should be noted that this formula is not additive or strictly multiplicative; instead it considers the variable characteristics in a manner supported by the theory.

The formulaic expression ⁷ is (Bressers 2005):

Predicted type of interaction = (M +) × (I +) × [1 − (M−) × (P−)]

	Where (M +) is degree of positive motivation of the positive actor, 8
	(I +) is completeness of needed information of the positive actor(s),
	(M−) is degree of negative motivation of negative actor, 9 and
	(P−) is the balance of power as viewed from the most negative actor where
	(0.0 = negative actor has no power)
	(0.5 = balanced power)
	(1.0 = negative actor has all power).

To better understand the potential of the CIT, we compare the formula values with the observed results. Unfortunately it is not possible to also produce intermediate scores for observed results in this study. To distinguish observed results we use a seven-point scale arranged from most to least cooperative. While unable to provide intermediate results in terms of outcome, the scale still denotes important degrees such as the distinction between active/passive cooperation and forced cooperation, or, for example, can distinguish between opposition and obstruction.

1.	Active/passive cooperation.
2.	Forced cooperation.
3.	Opposition.
4.	Learning towards (active/passive) cooperation.
5.	No interaction/learning towards (forced) cooperation.
6.	Obstruction.
7.	No interaction.

The formula provides a more difficult test for the theory's predictability potential, asking not only if all of the expected cooperation cases are observed as cooperation, but also whether the range of values for cooperation cases (representing the variations in interactions) connect meaningfully with case observations. In this study, observed results do not represent the same level of variability, instead existing on the seven-point scale.

Using the formula adds the implicit assumption that there are linear relationships between the dependent and independent variables. More specifically, while the original theory asks only whether the implementer has “sufficient” information or is “positively” motivated, the formula assumes that every degree of lack in information or every missing percentage point of full motivation leads to an equal decrease in expected outcome.

We test the hypothesis that there is a relationship between theory expectations and observations (that the theory has predictability potential). More specifically, we test for an inverse (negative) relationship, since the dependent variable has the highest numeric value for the most unfavorable situation. We find a correlation (n = 46), using Spearman's Rho correlation coefficient (Rho = −0.66, p < 0.000). While the figures for the independent variables are on a quasi ratio level, the figures for the dependent variable are not, making this ordinal level coefficient most appropriate. This more strenuous test of correlation produces a lower test statistic, and lower associated probability, but it remains well within the boundaries of our predetermined alpha level of 0.05. It is clear, though, that the differentiation of expected outcomes produces lower correlations than the almost absolute matching results of the case by case comparison that used the flowchart with distinct values for the independent variables.

A scatter plot of the data in Figure 2 provides insight into this relationship, allowing us to make several observations. Since the scale of the dependent variable starts with 1 for active/passive cooperation, higher predicted values connect with lower observation values and conversely lower numeric predictions merits higher numeric observations. This leads to an inverse correlation. As is also shown in the scatter plot, there is a great deal of variation among formula predicted scores for observed cooperation/active cooperation cases. ¹⁰ There is also a somewhat smaller range of values for cases observed as opposition. ¹¹

Figure 2. Scatter plot of the relationship between observed and expected results using the formula

Analysis Two: Discussion

The three cases of obstruction (6 on the y axis) in general seem to do a bit better than the “learning cases” (line 4 on the y axis) in which the process atmosphere is more cooperation oriented. Lacking information can obviously be as restricting as conflict. They also show variation and contain the most notable outlier. This case is the Boone Slough case where a strong motivation and full information of the implementer, coupled with a weakly negative motivation and a strong power position, but one that does not overwhelm that of the implementer, produces a high predicted score for the interaction. In reality, the reluctant landowner, having lost his interest in the project, simply blocked the project's progress. Obviously, he has a stronger veto power than the balance of power score suggests. Without this one case the correlation rises to Rho = −0.74 for the observed outcomes (Pearson's r would rise to r = −0.82). This example illustrates that the formula's assumption, that continuous values also have a continuous impact on the results, might be incorrect.

On the other hand, there is the case of Farnham Park, which was the only case that, when using flowchart analysis, the theory expectation did not match with observed reality. In this analysis it is quite “in line” (the fourth one in row 3). While the lack of negative motivation of the target led to a false expectation of cooperation and success, here the moderate positive motivation and information scores of the implementer leads to a modest 0.50 prediction that fits rather well.

As stated above, when both actors are positive, in this formula version we use not the information of the most motivated actor, but of the best informed actor. There are six cases in which this makes a difference, though typically not a big one. When the information score of the most positive actor is used instead, Rho decreases from −0.66 to −0.62. Using not the scores of “the most positive actor” but instead the scores of the actor we discerned as the implementing actor (and also this actor's own information), leads to changes in 11 cases, and a Rho declining to −0.56 (all significant on the 0.000 level). Therefore, these alternative analyses do not indicate that changing the formula specification is necessary.

Analysis Three: Results

Another point for analysis was the influence of the types of states on this analysis (i.e. liberal Anglo-Saxon states (New Jersey and Oregon) and social-democratic welfare states (The Netherlands, Finland)). We ask how these ideologies might influence the relationship between predicted and observed results: is the theory equally predictive for both types of ideologies? In analyzing American states, we find a correlated inverse relationship (n = 24), using Spearman's Rho correlation coefficient (Rho = −0.54, p < 0.003). ¹² For comparison we next analyze the subsample including Finnish and Dutch states; we find a correlated inverse relationship (n = 22), using Spearman's Rho correlation coefficient (Rho = −0.80, p < 0.000). The test statistic is slightly higher for the social-democratic welfare states in comparison to the liberal Anglo-Saxon states, but this can be fully explained by the absence of the outlier.

We next split the cases into high population density states (New Jersey, The Netherlands) and low population density states (Oregon, Finland) to ask whether the theory predictability potential hold true in both high and low population density states. In analyzing high population density states, we find a correlation of Rho = −0.71 (p < 0.000, n = 24). For comparison we assess low population density states, finding a correlation of Rho = −0.62 (p < 0.000, n = 22, without the outlier Rho –0.74), a significant inverse relationship. Both associated probabilities fall well below our pre-set alpha level of 0.05, therefore the contextual interaction theory remains highly predictive in both low and high population density states.

Analysis Three: Discussion

There is no reason to believe that the contextual interaction theory is less valid in more pluralistic American states than in consensus-oriented European welfare states or that it is less valid in low density states. In fact even the small samples of most separate states show similar results (including Oregon when the outlier is removed). Given the constraints of a small sample size, these results must be tested in other places, policy situations, and empirical fields to create a continuing understanding about the theory's limitations and possibilities. It should be noted that for each “significant” statistical test, the associated probability was actually well below the pre-set alpha level of 0.05.

Conclusions

This research uses the contextual interaction theory to understand how the actor characteristics of motivation, information, and power influence implementation processes. This analysis stretches the previous applications of the theory in a number of ways, including the changed setting of wetland restoration projects, and extensions to countries outside of the Netherlands such as Finland, and the American states of New Jersey and Oregon. This analysis builds a steady argument for the applicability of the CIT as an explanatory tool in a number of new situations. We find that this theory represents a straightforward, consistent instrument for analyzing implementation processes, allowing comparability and the ability to replicate research.

In testing the theory in a 48 comparative case analysis, we found that it produced a good fit between expected and observed results. This predictive capability provides support for the explanatory power of this theoretical tool. When comparing expected and observed results, the validity of the theory is highly supported. These analyses are encouraging for building general theory on an important research question that thus far has not had much general theory validated. We find that when applying this theory, a useful tool emerges to depict how actors implement. While context matters, comparative analysis and general theory are here shown to be feasible.

In the formulaic test, the number produced by the formula represents a theory-based expectation about how actors with given characteristics will interact during the implementation of that restoration. We compare this number to the observed policy interaction occurring in each case, using correlation to highlight relationships between the given variables. We found that when testing the whole sample and sub-sets using the formula, we produce a significant test of correlation. For each significant statistical test, the associated probability was actually well below the pre-set alpha level of 0.05. These results, though still impressive, did not produce an equivalent fit. This provides evidence that implementation in practice incorporates threshold values in the core explanatory factors, for example, levels of motivation that do spur people into negative or positive action as well as a broad neutral category that does not prompt action in this way.

This analysis provides evidence for the applicability of this analytical tool, as well as several areas to investigate. The theory provides insight into how actors implement and in practice can provide a roadmap allowing researchers to identify and eliminate the barriers to implementation. The theory also enables a format for comparative, replicable analyses: an ability to gain insight into implementation in general through more comparative analyses. In addition, the formula test indicates that substantiation will be necessary to explore the threshold values of the core characteristics. Finally, replication and further comparative applications are necessary to better understand the theory's utility. It will be important to test the theory in additional scenarios, for example in developing countries, with different policies, and in multi-actor and multi-institutional settings.

Acknowledgments

The authors would like to thank Laurence J. O'Toole for his valuable and supportive comments on an earlier version of this manuscript.

Notes

1. A debate Saetren (2005) credits with causing many researchers to flee implementation studies altogether.

2. For a full description of operationalization, please see Owens (2008).

3. For an in-depth description of project methodology see Owens (2008).

4. A more in depth explanation of scoring and scales is available as Appendix 2.

5. See Bressers (2005) for a thorough treatment.

6. When using only the formula version, one cannot see for example whether a positive outcome is achieved by active or forced cooperation, or whether a negative outcome is produced by lack of motivation, delay because of essential learning, or obstruction by powerful opponents. In contrast, when using the flowchart model, this is clear.

7. In essence this formulaic expression takes into account the motivation of the positive actor and that actor's information level. It multiplies this value by an expression that quantifies the motivation of the negative actor and that actor's power as deducted from the whole number 1. For each individual case in the sample this formula produces a positive number on a scale of 0.0 to 1.0.

8. The value used in the formula for M+ is on the scale of 0.0 to 1.0. If neither actor is positive, 0.0 will be used in the formula; a negative value is not possible for M +.

9. The M− value is the absolute value of the degree of negative motivation on a –1.0 to 0.0 scale, so −0.23 becomes 0.23 for M− in the formula, but if the “most negative” actor is positive (e.g. +0.23) M− becomes 0.0, as there is no real negative motivation. The [1−(M−) × (P−)] term in that case automatically becomes 1.0, implying that no harm to the outcome will be done by opposition.

10. Number 1 on the y axis

11. Number 3 on the y axis

12. This correlation is only a bit lower than the general one by the increased relative impact of the “outlier” Boone Slough. Without it the correlation is Rho −0.69, p < 0.000, n = 23.

Appendix 1. Predictions of process interactions corresponding with the flowchart

Table

Prediction number from flowchart	Prediction
9, 14	For any interaction to evolve, it is necessary that application of the instrument would contribute positively to the motivation of at least one actor.
2, 8	If application of the instrument would contribute positively to the objectives of one actor, while the other actor is also positive or neutral, but the information of the positive actor(s) is insufficient to apply the instrument, then a joint learning process will evolve that will sooner or later create another situation.
6, 13	If application of the instrument would contribute positively to the objectives of one actor, while the other actor is negative, and the information of the positive actor is insufficient, then there will initially be no interaction, but the positive actor will try to learn on its own and thereby to create another situation.
1, 7	If application of the instrument would contribute positively to the objectives of one actor, while the other actor is also positive or neutral, and the information of the positive actor(s) is sufficient to apply the instrument, then the interaction process will have the character of cooperation. When both actors are positive there will be active cooperation.
3, 12	If application of the instrument would contribute positively to the objectives of one actor, while the other actor is negative, and the information of the positive actor is sufficient, then the character of the interaction process will be dependent on the balance of power between the actors. Dominance of the positive actor will lead to (forced) cooperation.
5, 10	If application of the instrument would contribute positively to the objectives of one actor, while the other actor is negative, and the information of the positive actor is sufficient, then the character of the interaction process will be dependent on the balance of power between the actors. Dominance of the negative actor will lead to obstruction.
4, 11	If application of the instrument would contribute positively to the objectives of one actor, while the other actor is negative, and the information of the positive actor is sufficient, then the character of the interaction process will be dependent on the balance of power between the actors. A relatively equal balance of power will lead to opposition. Opposition can take the forms of negotiation and conflict.

Appendix 2. Description of scoring

Motivation: responses are given positive and negative points based on whether they reflect motivation for (+) or against (–) the project. The resulting score is the proportion of positive responses divided by the total number of responses on a scale of 0.0 to +1.0. To create scores of “negative” and “positive” distribution, (0.50) is subtracted from all original scores, changing the scale of (0.0 to +1.0) to (−0.50 to +0.50). Next, this (−0.50 to +0.50) scale is multiplied by 2 and transformed to a scale of (–1.0 to +1.0). Within this configuration, motivation can be thought of as:

	−1.00 to −0.21 = negative motivation
	−0.20 to +0.20 = neutral motivation
	+0.21 to +1.00 = positive motivation

This transforms interview results to follow a scale that fits the theory conceptualization.

Information: responses earn positive or negative points as they depict the level of information held by each actor. The interview score is based on responses indicating positive levels of information as a proportion of total number of relevant questions, on a scale of (0.0 to +1.0).

Power: exists as a proportion of responses indicating sources of power within the process over the total number of relevant questions on power for each actor. As with the information score, power exists on a scale of (0.0 to +1.0). Additionally, the two actor power scores are compared to each other to determine which actor holds the balance of power. Specifically, a difference of (0.0 to 0.14) between the two scores indicates the power is balanced between the actors; a difference of (0.15) points or greater indicates that one actor holds the balance of power over the other actor.