- Master Thesis - Diffusion of a changing template - An explorative study analyzing the spread of bioenergy villages in Germany

(1)

- Master Thesis -

Diffusion of a changing template - An explorative study analyzing

the spread of bioenergy villages in Germany

Student: Daniel Sturm

Student number: S3412415

E-Mail: d.sturm@student.rug.nl

University of Groningen

Faculty of Economics and Business

MSc. Business Administration – Change Management

Groningen, January 21st_{, 2019}

Supervisor: Dr. B.C. Mitzinneck

Co-assessor: Prof. Dr. ir. D.J. Langley

(2)

Abstract

Attempting to understand the factors affecting diffusion past research focused mostly on the impact of predictors on the speed of diffusion. Common predictors analyzed in this context are distance and time. To this day, research regarding the effect of geographical and temporal distance on the fidelity of a diffused practice is scarce. Utilizing quantitative methods, this thesis contributes to existing literature analyzing the impact of the above predictors on the fidelity of a diffused practice. Precisely, a hierarchical multiple regression analysis is used to study the influence of the predictors on fidelity. The analysis draws data from 147 bioenergy villages across Germany. This study reveals no significant relationship between neither geographical distance nor time and the fidelity of a diffused practice. A showcase model represents the initial state of the diffusing practice. The moderator, prior local example influences the relationship between the distance and the fidelity of the diffusing practice negatively and significant. Furthermore, an active support program, moderates the relationship between the time since the showcase model and the fidelity negatively. Both moderation effects are subject to cross-over interactions. Prior local examples as well as support programs have an amplifier effect on the analyzed relation. The discussion part of the thesis contains theoretical and practical implications, showing interesting aspects for future research.

Keywords: bioenergy, bioenergy villages, cross-over interaction, diffusion, distance, fidelity, modern

(3)

Introduction

1. Introduction

(4)

Theoretical foundation and Hypotheses Building moderating effect of an active support program regarding the impact of time on the fidelity of the diffused practice is evaluated. In summary, the research goal is to examine the effects of geographical and temporal distance on the fidelity of a diffusing practice. The following research question describes the presented field of interest:

What is the relationship between geographical and temporal distance on the fidelity of a diffusing practice? The open nature of the research question implies that possible results remain uncertain. Nevertheless, the outcome of this quantitative study will give an insight into the diffusion characteristics throughout the population of BEVs. This scrutinizes the impact of temporal aspects, as well as the influence of the local environment regarding the fidelity of a diffusing practice, filling a gap in modern diffusion theory. Therefore, this quantitative study will contribute to the analysis of the effect of geographical and temporal distance on diffusion. In case significant results are found, policymakers can derive implications in a way, that they can incorporate the gained knowledge about the impact of a showcase model on diffusion in respect to geographical and temporal factors. Furthermore, they can adjust the structure of their programs accordingly.

2. Theoretical foundation and Hypotheses Building

(5)

Theoretical foundation and Hypotheses Building

2.1 Classical diffusion theory

Rogers (1995) provides a compelling definition of diffusion, describing it as a process of an innovation being communicated through channels over time amongst members of a social system. This description depends on the assumption that practices, which are subject to diffusion are invariant and do not change throughout the diffusion process. Therefore the time of adoption has no impact on the diffused practice. The entity which is new to the practice can either accept or reject the practice, but not change it. In short, the practice is commonly accepted and taken for granted (Tolbert & Zucker, 1983) particularly in contexts rife with considerable uncertainty. The expectation is a simple emulation of a previous layout. Westphal, Gulati, and Shortell (1997) emphasize the invariability of the diffused practices, enhancing it with the fact, that practices are not contestable. They highlight, that especially for intrapopulation diffusion, spatial homogeneity is a pivotal assumption. Later, Strang & Soule (1998) add more emphasis on spread being a rather viscerally flow from source to an adopter. Classic diffusion theory is easily distinguishable in that the focus of the research is not only on whether but also how fast practices are diffused. Research is carried out on a broad spectrum of predictors. Webber (1972) for example analyzed the influence of location and time on diffusion, yet he did not challenge the assumption, that the diffused practice could be subject to adaptation. In conclusion, a diffused practice regarding classical diffusion theory, is uniform and inevitable, yet the speed of diffusion is influenced by a multitude of factors.

2.2 Diffusion of innovations

(6)

(7)

Theoretical foundation and Hypotheses Building implementation of the practice is crucial in understanding diffusion processes. Another critical factor to mention is the general focus of diffusion research. While many papers on classical diffusion target diffusion processes within companies, recent papers show a shift in focus allowing for an analysis of diffusion amongst companies (Fiss et al., 2012; Fiss & Zajac, 2004). Ultimately, modern diffusion theory constitutes the process of diffusion as neither uniform nor inevitable, contradicting or rather extending classical diffusion theory in many aspects regarding the adaptability and de facto adaptation of diffused practices.

2.3 A framework to measure diffusion

(8)

Theoretical foundation and Hypotheses Building in capturing both, the nature of the practice, as well as its actual implementation at the same time. It facilitates detailed analysis of practice variation.

2.4 The case of geographical and temporal distance

Temporal and geographical distance are often analyzed variables in diffusion research (i.e., Haynes, Mahajan, and White 1977; Mahajan and Peterson 1979; Webber 1972). Mostly, the analysis features a classical diffusion perspective, focusing on the impact of those variables concerning the speed of diffusion, rather than their impact on the diffusing practice. It is a common practice to analyze the two variables separately, distinguishing between their imminent impact on diffusion. It is less common to research them jointly. The analysis of local spillover effects however, with its inherent properties concerning time and distance, condenses that (Aldieri et al., 2018). Local spillover effects describe diffusing practices caused and influenced by spatial proximity.

2.5 Time of adoption

(9)

Theoretical foundation and Hypotheses Building offering a reference model. As external conditions change over time and small adaptations happen between adoptions, this results in a negative influence on the fidelity within the new BEV, as the characteristics vary more and more compared to the showcase model in later stages of the diffusion process. As the impact of time prevails, a larger timespan between the showcase model and a new venture negatively influences the fidelity of the new BEV, resulting in more different BEVs. Hence, based on the previously elaborated inferences, the following hypothesis derives:

Hypothesis 1: A larger timespan between the new venture and the showcase model will have a

negative effect on the fidelity of the implemented BEV.

2.6 Geographical distance

(10)

Hypothesis 2: A higher geographical distance will have a negative effect on the fidelity of the

implemented BEV.

2.7 Moderating effect of prior local example

The later an innovation is adopted over time, the more prior experience of others implementing that practice is available (Ansari, Fiss, and Zajac 2010). In the context of the research in this thesis, the experiences were collected and published in a guidebook by the German government. The underlying expectation is that aspirant communities not only look for guiding papers but seek further sources of information. One possible source of information an aspirant community may draw from could be a prior local example within the same district, resembling a source of experience from their founding process. In other disciplines, geographical spillover as mentioned by Aldieri, Kotsemir, and Vinci (2018) describes information being transferred due to the fact, that the entities are close to one another. Looking into this is of particular interest, to account for the influence another BEV, already existing in the same district of the aspirant community, has. In this case, a low disparity of geographical distance is a given, and the availability of a prior local example could influence the decision-making process of other communities thinking about adopting the BEV practice. A possible area of effect is uncertainty, which, when reduced, possibly leads to a better risk awareness resulting in overall better planning of the new venture (Wei et al., 2015). Based on the assumptions made on the influence of geographical distance before, the influence of a present prior experience in form of a local example, assumingly is higher, when the geographical distance to the showcase is higher, resulting in a reduction of uncertainty during the planning process, leading to an overall better fit of the implemented practice regarding the local needs. This negatively influences the fidelity within the new BEV, as the influence of the showcase model is lower, when another BEV exists in the vicinity. Therefore, if a BEV is present in the district when the new venture is founded, the fidelity is lower. Hypothesis 3 is testing this assumption:

Hypothesis 3: A prior local example negatively moderates the relationship between the distance to

(11)

2.8 Moderating effect of support programs

As formerly mentioned in the introduction of this thesis, the German government invests heavily into funding support programs, supporting aspirant villages on their way to becoming a BEV. Toka et al. (2014) research the impact of support programs on diffusion and found a significant impact. In the context of this thesis, their rather classical view on the impact of support programs on diffusion is extended analyzing the impact of support programs on the fidelity of a diffusing practice. Merely analyzing the predicting effect of support programs is insufficient, as the common goal of the related support programs is to give aspirant villages the necessary leeway to explore their options to the fullest (Heck et al., 2014). This enables them to maximize their benefits profiting from prior experiences made by other BEVs prior to making the step into energy independence. This could potentially lead to an overall better fit as described by Ansari et al. (2010). As the transmitted practice is adapted to fit the specific needs of the implementing village, resulting in local adaptations lowering the fidelity, as the practice grows more and more diverse compared to the showcase model. Overall, the influence of time on the fidelity of a diffusing practice is negatively moderated by an active support program, especially during later stages of the diffusion process. This leads to the fourth hypothesis:

Hypothesis 4: Active support programs negatively moderate the relationship between time passed

(12)

Methodology

3. Methodology

The methodology section provides an outline for the present study not only introducing the context of the study but also describing the data collection process used to gather the sample data used in the empirical analysis. Furthermore, the data measures are introduced, operationalized and discussed. The data analysis method is described, outlining the utilized model of hierarchical regression analysis.

3.1 Study Context

The present study aims to enhance existing literature offering a view on the influence of time and distance on the diffusion process, not as commonly done, on the speed of diffusion, but on the fidelity of the implemented practice. The analysis will be done, scrutinizing factors influencing the diffusion of bioenergy villages in Germany. In detail, this study analyzes the effect of time and geographical distance on the fidelity of a new BEV. The time of attaining the BEV status is of particular interest, as it implies, how much prior experience is present and how much time passed since the showcase model was introduced (Ansari et al., 2010; Stoneman & Diederen, 1994). Geographical distance is a decisive aspect in this context, as the existence of a showcase model and its effect on the diffusion of the practice could counteract the expected outcomes. Both variables are analyzed regarding their influence on fidelity, as part of a framework introduced by Ansari et al. (2010). The context is of particular interested because it combines research on diffusion between commonly researched influencing factors and their impact on fidelity (Ansari et al., 2010; Caiazza & Volpe, 2017).

3.2 Data collection

(13)

Methodology rebuilding process of the dataset, most information was drawn from the self-reports gathered by the FNR. Those reports were filed in the last years and contain data on BEVs founded between 2005 and 2016.

(14)

Methodology directly from the community administration. Completing the data from either the website or through direct request was necessary for 52 out of the 147 datasets.

Furthermore, additional data was drawn from different governmental databases, all maintained by the ‘Statistischen Ämtern des Bundes und der Länder’ (Federal Statistical Office). Namely the accessed databases are ‘Regionalstatistik’ (Regional Statistic) and ‘Statisitkportal’ (Statistic Portal). Minor amounts of information were gathered from ‘Komsis’ a local database founded by the state of Lower Saxony, specifically the ‘Institut für Regionalentwicklung’ (Institute for regional development). Finally, the author of this thesis gathered the data on village shape. After the categories for village-shapes were defined, based on the similarity to the showcase model, all 147 villages were searched on Google Maps, screenshots were taken and categorized according to their shape. To eliminate potential bias, the categorization process was conducted by the author of this thesis as well as an additional person, the results matched in 140 out of 147 cases, matching in 95% of the observations.

After a final screening for invalid data scores or missing data was conducted no observations were dropped due to missing values or other anomalies. The described process led to the complete dataset utilized for this thesis.

3.3 Measurement

The following section outlines the variables and their measurements. Starting with the dependent variable fidelity, followed by the independent variables geographical distance and time, concluded with the moderator variables support program and prior local example. Finally, the control variables are operationalized.

3.3.1 Dependent variable

(15)

Methodology can be measured using the two-dimensional framework of Ansari et al. (2010). This thesis follows one of the proposed variables, namely fidelity.

The dimension of fidelity analyzes, how comparable the local adaptation of the implemented practice is, in comparison to the other adaptations of the practice. Consequently, the implementer of the practice always needs to account for local circumstances, as well as internal requirements. Therefore, characteristics of the diffused practice may vary from observation to observation (Wu & Chiu, 2015). Fidelity is a suitable measure to analyze the differences between a showcase model and subsequent versions of it (Lewis & Seibold, 1993). Fizz et al. (2012) renamed fidelity into ‘similarity’ to clarify the term, as it analyzes how similar the different adaptations are. Keeping this in mind simplifies picturing the hypotheses. In order to put that into practice within the context of this thesis, a composite score for fidelity was calculated using the Euclidean distance between three factors. The three elements of the composite score are the primary resource used (i.e., biomass, wood), other resources (i.e., photovoltaics, hydroelectric power) used and the legal form of the venture of the BEV. The calculation follows the logic introduced by Doty and Glick (1994), calculating the Euclidean distance for three dimensions.

3.3.2 Independent variables

(16)

Methodology Following Katz (1999), diffusion studies should address time as a factor within the analysis. The way studies address time as a factor is alternating from study to study. The alternations happen due to the distinct emphasis a study has on time as an element of influence. Mullen, Nord, and Williams (2005) for example distinguish between fixed periods in time, whereas Grübler (1991) or Fiss et al. (2012) account for the specific years of the observation. Following this approach, this thesis will account for the time variable, utilizing a clock counting year wise. The year of the showcase project represents the zero-value counting up to the year the observed BEV obtained its status.

3.3.3 Moderator variables

A moderator analysis is utilized to assess the effect of a third variable on the relationship between a dependent and an independent variable. This thesis features the moderating effect of a prior local example on the distance and the effect of an active support program on the period since the showcase model.

Prior local example

(17)

Methodology

Support Program

Caiazza and Volpe (2017) explore a multitude of ways support programs can influence the diffusion of an innovation. Toka et al. (2014) research the impact of support programs in the context of bioenergy measures and find a significant impact. The German government invests heavily into funding support programs to assist aspirant villages on their way to achieving the BEV status. Following the indications of previous studies, the support programs should significantly impact predictors influencing the fidelity of new BEVs. In the context of this thesis, support program is measured in a categorical variable differentiating between an active support program at the time when the BEV status is achieved and no active support program at that time.

3.3.4 Control variables

A multitude of factors influences the diffusion of an innovation or an innovative practice. Wejnert (2002) categorizes three major components affecting the diffusion of innovations, split up according to their area of effect. The three categories of variables are, ‘characteristics of innovations,’ ‘characteristics of innovators,’ and ‘environmental context,’ In order to achieve a good fit of control variables, this thesis strives to cover all three categories by picking the most fitting control variables for each category.

(18)

Methodology The following set of variables controls for the characteristics of the innovators. Accounting for the characteristics of the innovators, the village shape, as well as the average percentage of votes for the German green party ‘Bündnis 90 die Grünen’ (i.e. green votes) over the observation period are analyzed. Considering the green votes is especially important, due to the fact, that political stability influences the adoption of innovations (Berry & Berry, 1990). Calculating the average green votes for all parliamentary elections within the observation period gives a conclusive insight over time. The analysis of the village shape compares the observations with the showcase model in Jühnde, allowing for valuable insights. It is split up into dummy variables accounting for different village shapes due to the categorical nature of the variable. ‘Village Shape 0’ indicates villages very similar to the showcase model, whereas ‘Village Shape 1’ indicates a slightly different village shape and ‘Village Shape 2’ indicates a significantly different shape. A village with the ‘Village Shape 0’ is circular, widespread and large. This also represents the shape of the showcase model. In case of ‘Village Shape 1’, the village is slightly different to Jühnde. Therefore, either huge (also supported by the number of inhabitants) or a non-circular stretched village alongside one road with some side roads. Whereas a village with ‘Village Shape 2’ is either a super small village with only a couple of households in a circular shape or a rather small village stretched alongside one road without major side roads. The shape criterions where chosen in regard to the ease of connecting the households with the local heat network.

(19)

Methodology ‘bioenergy region’ evaluates, whether the location of the observation is within an area declared as a bioenergy region and tests for consequential influences. Those regions were ‘active’ for certain periods in time, featuring support programs of different kinds regarding bioenergy in general, as well as particular programs for aspirant communities, on the way to become a BEV. Even if no longer active, the present experience in the area may influence the diffusion process, the general orientation, or rather mindset, of a region, can have a significant influence on the adoption decision of an aspirant village. Existing adopters provide experience and knowledge which can be utilized by followers (Grübler, 1991). A list of the support programs taken into account can be found in Appendix F. The difference between the support program variable and the bioenergy variable lies within the support program being active, whereas the bioenergy region is still there after the support programs runs out.

3.3.5 Data analysis

(20)

(21)

Results

4. Results

In this section, the descriptive statistics are laid out and the performed assumption tests are described. All underlying assumptions must be met to carry out a linear regression analysis. Furthermore, the presented hypotheses are tested, and a brief evaluation within the context is provided. Finally, two robustness checks are performed to eliminate potential bias.

4.1 Descriptive statistics

15.46

Cow density 83.23

182.5

1.627 1,078

Forest area 431.6

198.0

15.01 1,093

Preexisting BEV 0.367

0.484

0

1

Village shape 0 0.184

0.389

0

1

0.480

0

1

0.377

0

1

Table 1: Descriptive statistics (n=147)

(22)

Results Therefore, villages very similar and very different to Jühnde could influence the data over-proportionally. Performing the following correction accounts for the negative skewness and normalizes the data.

𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠2= 1/�((max(𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓) + 1) − 𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓2

(23)

(24)

Results

4.2 Assumption Checks

The following section will analyze, whether a multiple hierarchical regression analysis is a suitable quantitative method to test the hypotheses developed above. In order to verify the suitability of the method of analysis, the variables and data must meet certain criteria. In the case of a hierarchical regression analysis, it is necessary that the data complies with seven assumptions. To account for their focus, the assumptions are split into two groups. The first two assumptions have the objective to test the choice of study design and the utilized measurements. It is necessary to comply with all the latter five assumptions; this verifies that the data fit the regression model.

(25)

(26)

Results normal distribution. However, using the visual method can easily be flawed, as skewness is not easily observable from the visual inspection of the provided graph. Therefore, this thesis will analyze the normal quantile plot additionally. Even though this is a visual inspection as well; it is much easier to interpret. The normal quantile plot does show anomalies. Therefore, suitable corrective measures have been taken to adjust for the skewness. The descriptive statistics paragraph includes the exact description of the mathematical procedure. After correction, the normal quantile plot does not show any anomalies, as the curves are superimposed normal curves (see Appendix B). Therefore, it is certain that the data for all four hypothesis meets the assumption of normally distributed residuals. To rule out any margin of error the assumption tests are carried out per hypothesis. Appendix C includes the summarized results.

4.3 Hypotheses Testing

(27)

Results

VARIABLES Model 1 Model 2 Model 3 Model 4 Model 5

Independent variables Distance -0.012 -0.011 0.007 -0.012 (0.010) (0.010) (0.012) (0.010) Time 0.004 0.006 0.009 (0.005) (0.005) (0.006) Moderator variable Preexisting BEV#Distance -0.048*** (0.018) Support program#Time -0.029** (0.013) Support program 0.062 0.054 0.038 0.034 0.202* (0.039) (0.039) (0.044) (0.043) (0.086) Preexisting BEV 0.004 0.000 -0.008 0.102* -0.017 (0.022) (0.023) (0.024) (0.048) (0.028) Control Variables Green votes 0.004 0.004 0.004 0.003 0.004 (0.003) (0.003) (0.003) (0.003) (0.003) Cow density 0.000 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) Forest area 0.000 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) State BW -0.066 -0.047 -0.046 -0.015 -0.043 (0.042) (0.045) (0.045) (0.045) (0.044) State Bavaria -0.053 -0.048 -0.044 -0.043 -0.068 (0.093) (0.093) (0.094) (0.091) (0.093)

State Lower Saxony -0.016 -0.017 -0.013 -0.005 -0.029

(0.061) (0.061) (0.061) (0.060) (0.061) Bioenergy region -0.077** -0.077** -0.066** -0.064* -0.053 (0.030) (0.030) (0.033) (0.033) (0.033) Village shape 0 0.010 0.012 0.011 0.016 -0.054 (0.034) (0.034) (0.034) (0.033) (0.033) Village shape 1 -0.011 -0.008 -0.010 -0.006 -0.003 (0.028) (0.029) (0.029) (0.028) (0.024) Constant 0.820*** 0.842*** 0.821*** 0.764*** 0.804*** -0.047 -0.050 -0.056 -0.059 0.056 Observations 147 147 147 147 147 R-squared 0.0970 0.1080 0.1120 0.1564 0.1434 Adj. R-squared 0.0234 0.0276 0.0248 0.0669 0.0525 F (12, 134) 1.32 1.35 1.29 1.75 1.58

Standard errors in parentheses +0.1, *** p<0.01, ** p<0.05, * p<0.1

(28)

Results

4.3.1 Hypothesis 1

Hypothesis 1 predicts that a larger timespan between the new venture and the showcase model will have a negative effect on the fidelity of the implemented BEV. This prediction implies that the beta coefficient is expected to be significant and negative. The analysis of Model 2 reveals that there is no significant relationship between the fidelity and the timespan between the showcase model and the observation. Concludingly, there is no data to support hypothesis one. The hypothesis is rejected. Interestingly, the regression table shows an increase of 1.1% in the explained variance compared to model 1.

Hypothesis 2 predicts a negative relation between geographical distance and fidelity. To support the hypothesis, the beta coefficient is expected to be significant and negative. However, analyzing Model 3 shows, that this is not the case. Therefore, the expectation is disproved. This is highly counterintuitive to the expectations of an unbiased observer. The increase of the R2 _{by 0.4% compared to Model 2 shows an} increase in the explained variance. Nevertheless, there is no evidence to support hypothesis 2. Therefore, hypothesis 2 is rejected.

(29)

Results aforementioned interaction (i.e., Appendix D). Analyzing the simple slopes graph reveals a cross-over interaction. For observations close to the showcase, the distance has a significant influence, and therefore a prior existing BEV in the same district has no significant impact. This leads to observations being very similar to the showcase, when within close range. For observations with high distances from the showcase (after the intersection of the graphs), a prior local example (i.e., preliminary existing BEV) has an amplifying effect on the interaction strongly influencing it, leading to much lower fidelity versions of the diffused practice, observations being very different to the showcase. Additionally, analyzing the explained variance an increase by 4.4% to a total R² of 15.6 can be observed. It is important to mention that the crossover interaction shows that when a prior local example is present, distance has an opposed significant effect on the fidelity regarding the values right and left of the intersection point of the simple slopes graphs. Therefore, this interaction hides the significant effect of the predictor on fidelity as tested in H2.

The fourth hypothesis predicts that a BEV founded within a support region while the support program is active shows a negatively moderated relationship between the time since the showcase model and the fidelity of the diffused practice. BEVs founded within a region with an active support program are less similar to the showcase model, than BEVs which are not founded in a region with an active support program. Possibly, this is may be due to support programs providing financial leeway enabling aspirant villages to explore options optimally and adapt the diffusing practice to their local needs. To support the presented hypothesis the interaction term of this moderation needs to be significant with a negative coefficient. Model 5 shows that this is the case with a coefficient of -0.0286444 and a significance level of p ≤0.05. Analyzing the main effects, the indicator for an active support program is significant (p ≤0.05), yet the time indicator is not significant. Concludingly, there is evidence to support hypothesis 4.

(30)

Results The graph reveals a weaker impact of an active support program closer to the showcase compared to later points in time. The intersection is after seven years from the showcase, indicating that from there on an active support program has an amplifying effect resulting in much lower fidelity and villages being very different to the showcase model. Putting those findings into context, it indicates that, especially in the early stages of the analyzed period, the governmental support programs feature the showcase model as a favorable model. Therefore, BEVs in early stages are more similar to the showcase, when they are within an active support region, due to the influence of the support program. However, in the later stages, the advice provided did not necessarily change, yet the financial leeway the support programs provide enables the aspirant villages to explore local adaptations from close BEVs as well and adjust them for local needs, resulting in more different (higher fidelity) versions of BEVs to those in early stages of the diffusion process. Furthermore, it is worth noting that including the moderation effect increases the explained variance by 2.2 to a total R² of 13.4, compared to model 3. Furthermore, it is important to mention, that when a support program is present time has a significant effect on fidelity. Yet, the effect is opposing on the left and right side of intersection of the simple slopes graphs hiding a significant effect. This results in H1 to be not supported even though there is a significant relation between time and fidelity, at least when a support program is active. To summarize the findings, Table 4 provides an overview of the findings regarding all hypotheses.

Table 4: Hypotheses overview

Hypotheses Expectation Result

H1 The time between the new venture and the showcase model

negatively affects fidelity. Rejected

H2 Geographical distance negatively affects fidelity Rejected

H3 A prior local example negatively moderates the relationship

between the distance to the showcase model and the fidelity. Supported

H4 Active support programs negatively moderate the relationship

(31)

Results

4.4 Robustness Test

During the formation process of this thesis, the assumptions for the different hypotheses were tested. It appeared that fidelity is strongly negatively skewed. Therefore, the data was transformed according to the process described in the descriptive statistics section. The transformation utilized a mathematical construct and successfully reduced the skewness significantly. It is critical to test, whether the transformation had effects on the results of the regression analysis. Hence, to check the robustness, the same analyses which was done with the transformed value were performed with the non-transformed value for fidelity. It is expected that results will not differ between the analysis with transformed and non-transformed data. Beforehand, additional assumption tests were performed for the non-non-transformed data to ensure that there are no anomalies. This was the case, as all other assumptions for the chosen statistical method were met, confirming the appropriateness of the method of analysis. Results of the analysis can be found in Table 5. Corresponding with the analysis carried out with the transformed values for fidelity, neither the time nor the distance to showcase produces a significant relationship with fidelity utilizing the untransformed values. Analyzing the results of the regression analysis carried out for hypothesis 3 (Model 4), slight differences attract attention. Neither of the two main effects of the observed moderation is significant, yet the moderation effect itself is still significant with p ≤0.05. This represents an even more convincing result, as there were indications for a crossover-interaction beforehand. If neither of the two main effects is significant, this strongly suggests a crossover-interaction. Analyzing Model 5, testing the premises of hypothesis 4 the observation does not differ at all from the observations made analyzing the transformed values. The moderation effect is statistically significant with a p ≤0.05. The main effect support program is significant with a p ≤0.01 and the other main effect, time is not significant, indicating a crossover-interaction. The predictive capacities show a similar pattern to those observed in the transformed values and are therefore not further elaborated.

(32)

(33)

Results

VARIABLES Model 1 Model 2 Model 3 Model 4 Model 5

Independent Variables Distance -0.038 -0.035 0.028 -0.036 (0.043) 0.043 (0.054) 0.043 Time 0.013 0.019 0.033 (0.024) (0.024) (0.024) Moderator Variables Preexisting BEV#Distance -0.163* (0.082) Support program#Time 0.122*** (0.058) Support program 0.350** 0.325* 0.277 0.263 0.998** (0.172) (0.174) (0.196) (0.194) -(0.381) Preexisting BEV -0.022 -0.034 -0.057 0.315 -0.003 (0.100) (0.100) (0.109) (0.217) (0.024) Control Variables Green votes 0.019 0.021 0.021 0.018 0.017 (0.015) (0.016) (0.016) (0.016) (0.015) Cow density 0.000 0.000 0.000 0.000 -0.000423* (0.000) (0.000) (0.000) (0.000) (0.000) Forest area 0.000 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) State BW -0.297 -0.236 -0.234 -0.129 -0.290 (0.187) (0.199) (0.200) (0.205) (0.182) State Bavaria -0.297 -0.282 -0.269 -0.264 -0.388 (0.416) (0.417) (0.419) (0.414) (0.414)

State Lower Saxony -0.098 -0.102 -0.089 -0.064 -0.155

(0.271) (0.272) (0.273) (0.271) (0.270) Bioenergy region -0.372*** -0.374*** -0.340** -0.333** -0.283* (0.135) (0.136) (0.149) (0.148) (0.148) Village shape 0 0.121 0.127 0.126 0.142 0.117 (0.152) (0.152) (0.152) (0.151) (0.150) Village shape 1 -0.018 -0.009 -0.015 -0.001 -0.055 (0.127) (0.127) (0.128) (0.127) (0.126) Constant 1.073*** 1.141*** 1.076*** 0.882*** 0.939*** -0.207 -0.221 -0.251 -0.267 -0.229 Observations 147 147 147 147 147 R-squared 0.110 0.115 0.117 0.117 0.142 Adj. R-squared 0.037 0.035 0.030 0.051 0.063 F (12, 134) 1.32 1.35 1.29 1.75 1.58

Standard errors in parentheses +0.1, *** p<0.01, ** p<0.05, * p<0.1

(34)

Discussion and Conclusion

5. Discussion and Conclusion

Describing the perks of classical diffusion theory leading over to modern diffusion theory and finally analyzing the specificities of fidelity, this thesis analyses the impact of multiple predictors on the fidelity of a diffusing practice. Therefore, the main research question scrutinizing the impact of geographical and temporal distance on fidelity was developed, hypotheses were laid out, and tested. In short, it can be captured that there is no significant relationship between neither geographical nor temporal distance and the fidelity of a diffusing practice in the researched context. In the following paragraphs, the theoretical, as well as the practical implications of those findings, are laid out. Furthermore, the limitations and implications for future research are discussed.

5.1 Theoretical contribution

(35)

Discussion and Conclusion theory. So far, mostly classical diffusion theory papers like Mahajan and Peterson (1979), analyze the impact of a showcase model. This thesis tries to elaborate on the effects of a showcase from a modern diffusion theory perspective. It is important to mention that despite the fact that H1 and H2 are not supported the analysis of H3 and H4 shows significant relations for both respectively researched predictors within the researched interaction. The analysis of the interaction shows, that the researched predictors have a significant effect on fidelity, yet because they are subject to a crossover interaction the significant effect on fidelity is hidden, when not researched in an interaction term. Future studies can benefit tremendously from this knowledge in a way that if they are researching predictors influencing the fidelity of a diffusing practice and they find insignificant results, they can look into interactions with other variables hiding a possible significance. Especially in a context with a showcase model those interactions could play a substantial role regarding the significance of certain predictors.

Summarizing, this paper enhances our understanding of the impact of time and distance on the fidelity of a diffusing practice utilizing the lens of modern diffusion theory. Enhancing literature, offering a good start for the analysis of the impact of showcase models in diffusion theory from a modern perspective, regarding their influence on practice variation, with the help of moderation effects.

5.2 Practical implications

This thesis allows both context-specific and overarching implications. The context-specific ones are at least in the short term much more important. The basis for most of the data utilized in this thesis are the self-assessment forms gathered by the FNR. So far, the FNR has no common database containing the combined information on all BEVs. Most likely due to the extensive process necessary to combine the data from the different self-assessment forms. Therefore, the data sheet developed for this thesis will be provided to the FNR alongside this paper for future use.

(36)

Discussion and Conclusion either more research needs to be conducted as the findings are counterintuitive to an unbiased observer or it can merely be used as an indication not to focus on those factors.

An exciting finding from the analysis is that fidelity is negatively correlated with bioenergy region. This is highly counterintuitive as the overarching concept of the German government and its organizations was to spread the showcase model as a non-adapted diffusing practice. Resources provided to aspirant villages as well as local transfer of information, spillover effects, seem to have a significant impact and need to be taken into account in the future. Possible reasons for this could lie within the properties of necessary practice adaptation to fit local needs. This knowledge can be utilized to alter the approaches taken in the support programs or the incentives provided, therewith either enhancing the leeway provided to aspirant villages or reinforcing the local adaptation efforts to achieve the best fit for each village or to restrain the same to feature the spread of a non-adopted version of the showcase.

On a general level, one can derive information about the interaction between a showcase model and support programs. The gathered information could be especially helpful in the area of change management as oftentimes great implementation processes, which are accompanied by change managers, are pioneered by one of many entities operating as a prototype or in the figurative sense as a showcase model. This thesis shows that there are moderating effects, which need to be taken into consideration, whenever a practice based on a showcase model, is diffused to multiple entities, as they are affecting the fidelity of the future adaptations.

5.3 Limitations

(37)

Discussion and Conclusion Another limitation is the imprecision of variables. The time variable in the context of this study is measured in years due to the availability of data gathered by the FNR. Acquiring the exact date, when the BEV achieved its status, would increase the accuracy of this variable and respectively the results of the study. Furthermore, the data for the variables measuring the political sentiment, the cow density, and the forest area is gathered on the district rather than the community level. Even though they should not diverge from one another too much on average, it is more accurate to gather the data on a community level. Due to the limited time available, this idea was rejected for this thesis. The data utilized for this thesis was gathered mostly from governmental databases which report on a district level.

Furthermore, there are minor limitations regarding the interpretability of results. The R² of Model 4 & 5 is 0.1564 and 0.1434 respectively. To enhance the explained variance of the findings more control variables can be explored to increase the amount of explained variance featured within the study.

Regarding the interpretability of the presented results, it is important to mention that in the results of this thesis correlation does not imply causation, due to the nature of the cross-sectional data analysis utilized.

5.4 Future research

The quantitative analysis of this thesis utilizes parts of the framework Ansari et al. (2010) propose in their paper to analyze practice variation during diffusion processes. Precisely, this thesis features one of the two dimensions utilized. The other variable explored by them is ‘extensiveness,’ measuring the degree of practice implementation within the new entity. To further elaborate on the findings of this thesis the relations found could be explored to a greater detail analyzing, whether they can be verified regarding extensiveness as well. Additionally, Ansari et al. (2010) propose to explore external factors affecting diffusion. Even though this thesis features some aspects regarding this topic; they are not explored to a satisfying detail offering the potential for further research.

(38)

Discussion and Conclusion within this thesis regarding support programs is whether there was one in the respective area active at the time of BEV status achievement. The derivation from the literature of hypothesis 2 mentions the impact of a showcase model. Implicitly this contains the impact of support programs as the support programs in this context featured the showcase model. Therefore, those effects might impact the effect of distance on the fidelity of the diffused practice. Further research is needed to verify this assumption. Additionally, it would be interesting to distinguish between different support programs and their respective goals. The target group could be a possible aspect. During the research for this thesis, it became apparent that support programs can target the community as well as the individual household. The different target group could influence the impact of the respective support program on the fidelity of the diffusing practice.

(39)

Discussion and Conclusion results. It is also important to notice the negative correlation of fidelity to the bioenergy region indicator. The observed correlation is counterintuitive as outlined in the practical implications part of this thesis. Additionally, the models 4 & 5 show a marginal (4) and a non-significant (5) relation to fidelity. As mentioned in the limitations sections, correlation in this thesis does not imply causation, due to the cross-sectional data utilized. Therefore, future research is needed to test for a possible causation.

Finally, there are few studies like the present one, exploring the impact of predictors on the fidelity of a diffusing practice in a context, where a showcase model is present. Further research is needed to explore the influence of showcase models on predictors influencing the fidelity of a diffusing practice.

5.5 Conclusion

(40)

Bibliography

6. Bibliography

Aldieri, L., Kotsemir, M., & Vinci, C. P. (2018). The role of geographic spillovers in employment policyplanning: an empirical investigation for Russian regions. Foresight, 20(3), 289–311.

Ansari, S. M., Fiss, P. C., & Zajac, E. J. (2010). Made to fit: How practices vary as they diffuse. Academy of Management Review, 35(1), 67–92.

Ansari, S., Reinecke, J., & Spaan, A. (2014). How are practices made to vary? Managing practice adaptation in a multinational corporation. Organization Studies, 35(9), 1313–1341.

Armstrong, J. S. (2012). Illusions in regression analysis. International Journal of Forecasting, 28(3), 689– 694.

Berry, F. S., & Berry, W. D. (1990). State Lottery Adoptions as Policy Innovations : An Event History Analysis. American Political Science Association, 84(2), 395–415.

Brandyberry, A. A. (2003). Determinants of adoption for organisational innovations approaching saturation. European Journal of Innovation Management, 6(3), 150–158.

Bundesministerium für Bildung und Forschung. (2018). Bundesbericht Forschung und Innovation 2018. Caiazza, R., & Volpe, T. (2017). Innovation and its diffusion: process, actors and actions. Technology

Analysis and Strategic Management, 29(2), 181–189.

Canato, A., Ravasi, D., & Phillips, N. (2013). Coerced practice implementation in cases of low cultural fit: cultural change and practice adaptation during the implementation of six sigma at 3M IÉSEG School of Management, 56(6), 1724–1753.

Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2013). Applied multiple regression/correlation analysis for the behavioral sciences. Routledge.

Cook, R. D., & Weisberg, S. (1982). Residuals and influence in regression. New York: Chapman and Hall. Doty, D. H., & Glick, W. H. (1994). TYPOLOGIES AS A UNIQUE FORM OF THEORY BUILDING: TOWARD IMPROVED UNDERSTANDING AND MODELING. Journal of Sustainable Development Studies, 19(2), 230–251.

Ekdale, B., Singer, J. B., Tully, M., & Harmsen, S. (2015). Making Change : Diffusion of Technological , Relational , and Cultural Innovation in the Newsroom. Journalism & Mass Communication Quarterly, 92(4), 938 –958.

Fiss, P. C., Kennedy, M. T., & Davis, G. F. (2012). How Golden Parachutes Unfolded: Diffusion and Variation of a Controversial Practice. Organization Science, 23(4), 1077–1099.

(41)

Bibliography Fiss, P. C., & Zajac, E. J. (2006). The symbolic management of strategic change: Sensegiving via framing

and decoupling. Academy of Management Journal, 49(6), 1173–1193.

Gondo, M. B., & Amis, J. M. (2013). VARIATIONS IN PRACTICE ADOPTION : THE ROLES OF CONSCIOUS REFLECTION AND DISCOURSE, 38(2), 229–248.

Gort, M., & Klepper, S. (1982). Time Paths in the Diffusion of Product Innovations. The Economic Journal, 92(367), 630.

Grübler, A. (1991). Diffusion. Technological Forecasting and Social Change, 180(39), 159–180.

Haynes, K. E., Mahajan, V., & White, G. M. (1977). Innovation diffusion: A deterministic model of space-time integration with physical analog. Socio-Economic Planning Sciences, 11(1), 25–29.

Heck, P., Anton, T., Böhmer, J., Huwig, P., Meisberger, J., Menze, S., … Wangert, S. (2014). Bioenergiedörfer. Gülzow-Prüzen: Fachagentur Nachwachsende Rohstoffe e. V. (FNR).

Hyysalo, S., Johnson, M., & Juntunen, J. K. (2017). The diffusion of consumer innovation in sustainable energy technologies. Journal of Cleaner Production, 162, S70–S82.

Katz, E. (1999). Theorizing Diffusion : Tarde and Sorokin Revisited. Annals of the American Academy of Political and Social Science, 566(1), 144–155.

Koenker, R. (1981). A note on studentizing a test for heteroscedasticity. Journal of Econometrics, 17(1), 107–112.

Lewis, L. K., & Seibold, D. R. (1993). Innovation Modification during Intraorganizational Adoption. Academy of Management Review, 18(2), 322–354.

Lounsbury, M. (2001). Lounsbury, 2001, Institutional sources of practice variation, ASQ. Administrative Science Quarterly, 46(1), 29–56.

MacVaugh, J., & Schiavone, F. (2010). Limits to the diffusion of innovation: A literature review and integrative model. European Journal of Innovation Management, 13(2), 197–221.

Mahajan, V., & Peterson, R. A. (1979). Integrating time and space in technological substitution models. Technological Forecasting and Social Change, 14(3), 231–241.

Mullen, J. K., Nord, S. E., & Williams, M. (2005). Regional skill structure and the diffusion of technology. Atlantic Economic Journal, 33(1), 115–131.

Pandis, N. (2016). Multiple linear regression analysis. American Journal of Orthodontics and Dentofacial Orthopedics, 149(4), 581.

Peres, R., Muller, E., & Mahajan, V. (2010). Innovation diffusion and new product growth models: A critical review and research directions. International Journal of Research in Marketing, 27(2), 91– 106.

Rogers, E. M. (1995). Diffusion of innovations. 4th Edition. Newyork Free Press.

(42)

Bibliography 104(425), 918.

Strang, D., & Soule, S. A. (1998). Diffusion in Organizations and Social Movements: From Hybrid Corn to Poison Pills. Annual Review of Sociology, 24(1), 265–290.

Strang, D., & Tuma, N. B. (1993). Spatial and Temporal Heterogeneity in Diffusion, 99(3), 614–639. Toka, A., Iakovou, E., Vlachos, D., Tsolakis, N., & Grigoriadou, A. L. (2014). Managing the diffusion of

biomass in the residential energy sector: An illustrative real-world case study. Applied Energy, 129, 56–69.

Tolbert, P. S., & Zucker, L. G. (1983). Institutional Sources of Change in the Formal Structure of Organizations: The Diffusion of Civil Service Reform, 1880-1935. Administrative Science Quarterly, 28(1), 22.

Webber, J. M. (1972). Impact of uncertainty on location. Cambridge, Mass: The M.I.T. press.

Wei, J., Lowry, P. B., & Seedorf, S. (2015). The assimilation of RFID technology by Chinese companies: A technology diffusion perspective. Information and Management, 52(6), 628–642.

Wejnert, B. (2002). Integrating Models of Diffusion of Innovations: A Conceptual Framework. Annual Review of Sociology, 28(1), 297–326.

Westphal, J. D., Gulati, R., & Shortell, S. M. (1997). Customization or Conformity? An Institutional and Network Perspective on the Content and Consequences of TQM Adoption. Administrative Science Quarterly, 42(2), 366.

White, H. (1980). A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity. The Economic Journal, 48(4), 817–838.

Whitten, P., & Collins, B. (1997). The Diffusion of Telemedicine Communicating an Innovation. Science Communication, 19(1), 21–40.

(43)

Appendix

7. Appendix

Appendix A: Variance inflation factors

Variable VIF 1/VIF

Support Program 4.28 0.235587 Bioenergy Region 2.85 0.350311 State Baden-Württemberg 2.13 0.469183 Village shape 1 1.96 0.510929 Village shape 0 1.82 0.550948 Time 1.70 0.588247 Preexisting BEV 1.45 0.691796 Distance 1.28 0.778882

State Lower Saxony 1.28 0.782446

Average green votes 1.27 0.786753

State Bavaria 1.23 0.816257

Forest area 1.20 0.833281

(44)

Appendix

Pre-correction:

After correction:

(45)

Appendix

Appendix C: Summary Assumption Testing

Assumption H1 H2 H3 H4

A1 Ok Ok Ok Ok

A2 Ok Ok Ok Ok

A3 Ok Ok Ok Ok

A4 Ok Ok Ok Ok

A5 (Scatterplot) Unclear Unclear Unclear Unclear

A5 (Studentized) Ok Ok Ok Ok

A5 (Cook’s D.) Anomalies Anomalies Anomalies Anomalies

A6 (Graph) Unclear Unclear Unclear Unclear

A6 (Breusch-Pagan) Ok Ok Ok Ok

A6 (White) Ok Ok Ok Ok

A7 (normal) Skewed (-) Skewed (-) Skewed (-) Skewed (-)

(46)

Appendix

Appendix D: Simple Slopes Graph hypothesis 3

(47)

Appendix

Appendix F: Support Programs

Figure 2: Bioenergy Regions Germany (FNR) Light Green: 2009-2015