and Adoption of Computer Applications Softwear in Local Governments.&dquo; This project ... business and industry (Mansfield, 1971; Utterback, 1974). Even if one examines .... efficiency, or to the extent that managers are optimistic about.
Adoptability is the probability that an innovation will be incorprated by an organization. This study explores the question of the adoptability of an innovation for a particular class of technological innovations, computer applications, within the context of American local governments. Four conceptual dimensions are explored to determine their independent and interactive influences upon the adoptability of an innovation.
likely
THE ADOPTABILITY OF INNO VA TIONS An Empirical Assessment of
Computer Applications in Local Governments JAMES L. PER R Y JAMES N. DANZIGER
University of California,
Irvine
studies of innovations has grown, has been expressed about the unstable, noncumulative nature of the findings. These critiques note that part of the problem is that the research lumped under the rubric of &dquo;innovation studies&dquo; encompasses an array of different theoAs the number of
increasing
empirical
concern
AUTHORS’ NOTE: The authurs wish to tlrank Irwin Feller for his helpful comments on earlier draft of this article. This article is part o/’a research project entitled diffusion and Adoption of Computer Applications Softwear in Local Governments.&dquo; This project is supported by a grant to the Public Policy Research Organization and the Graduate School of Administration from the Division of Policy Research and Analysis of the National Science Foundation (PRA76-15549). The views expressed herein are those of the researchers and should not be ascribed to the .National Science Foundation. This is a revised version of material appearing in James L. Perry and Kenneth L. Kraemer. Technological Innovation in American Local Government: The Case of Computing, published by Pergamon Press. an
461
462
retical
and different
approaches (Downs and Mohr, 1976; Yin et al., 1976; Warner, 1974; Havelock, 1969). In this study, we explore one of the dominant theoretical questions in these studies-the adoptability of an innovation. The adoptability question asks: What characteristics seem to affect the likelihood that one innovation or another will be adopted by an organization? Our empirical analysis of adoptability focuses upon a particular class of technological innovations, the use of computerized applications to accomplish various information processing tasks.’ The context of our analysis is a particular class of organizational units, American local governments. concerns
THEORETICAL FRAMEWORK THE UNIT OF ANALYSIS
dependent variable in this study is &dquo;adoptability,&dquo; which is the probability that an innovation will be incorporated by an organization. A useful conceptual refinement for the study of adoptability is central to the &dquo;innovation-decision design,&dquo; proposed by Downs and Mohr (1976: 706). They note that the adoption of an innovation is contingent upon its organizational context, and hence the analysis should be based on interactive measures, in which the unit of analysis is &dquo;the organization with respect to a particular innovation&dquo; or &dquo;the innovation with respect to a particular organization.&dquo; They observe: The
we were studying the adoption of 10 innovations by 100 organizations, we would be working with a sample of 1,000. This design eliminates any confusion that might stem from volatile secondary attributes.... A benefit of employing the innovationdecision design is that it serves to remind us of the dangers of thinking in terms of something called the organization, a reification with constant properties whose probability of adoption
If
varies, regardless of the kind of innovation it considers. // focuses OMr attention ~f~tOA! on o~ the ~~ shifting Rather it ~o~~y /~c~/h~ and at /bcM~~ our ~/!~/~ incentives constraints that are relevant to the decision to innovate [Downs and Mohr, 1976: 706; italics added]. never
463
While this study does not undertake a precise application of the design, it does conceptualize and measure the relevant variables as
contingent phenomena.
The data for this study of adoptability are primarily derived from the recent URBIS survey of local government computer use. More than 70% of the larger municipal and county governments (populations greater than 50,000 and 100,000 respectively)
mail questionnaire. Each government specified which of 261 automated applications in 26 functional areas were currently automated (in 1975). From this list of 261 applications we selected 10 which are treated as the innovations in this analysis. The key criterion employed for selecting these particular applications was to maximize variation in the diffusion patterns of the applications. This assured a diverse, broadrange sample of applications, and it also served to maximize variation in the attributes of the applications.2 Given the 10 applications and the approximately 350 governmental units for which there were data, our analysis employs an N of about 3500 cases (i.e., 10 innovations x 350 organizations).
responded
to a
DETERMINANTS OF ADOPTABILITY so many different sets of independent variables are innovation research, selection and operationalin employed ization of the variables meant to explain variation in adoptability are problematic. An abundant literature focuses on innovation in business and industry (Mansfield, 1971; Utterback, 1974). Even if one examines only the research on public agencies at the local level, the alternatives are extensive. The study by Yin et al. (1976), which aimed to examine and codify the many empirical findings on state and local government innovation, underscores the diversity of explanatory variables. Recent research by Feller and Menzel (1976, 1975) and by Bingham (1976) also identifies arrays of variables relevant to the adoption of innovations by local government organizations. Several recent studies have also explicitly examined variables associated with the adoption of computer applications by local governments (Perry and Kraemer,
Because
464
1978; Danziger and Dutton, 1977). Appendix A presents a list of
independent
variables from the innovation research that
seem
relevant to our research problem. While we are guided by the variables identified in the other research, we replace the many specific variables with a smaller number of more abstract dimensions. Employing a small number of general concepts in the study of adoptability has several advantages. First, using a small number of abstract dimensions facilitates the exploration of contingent relationships among key variables. For example, several studies suggest the plausibility of interaction among such variables as motivation, resources, executive ideology, and autonomy (Mohr, 1969; Downs, 1976). However, these types of contingent relationships become extremely difficult to isolate when one analyzes a large number of collinear variables. Limiting the analysis to a small number of dimensions facilitates the isolation and exploration of such contingent relationships. A second advantage of using a small number of general concepts is that it lessens the technical problems of complex interaction that arise when a large number of variables is used to study innovation (Downs, 1976: 130). Consequently, we have organized the independent variables in terms of four conceptual dimensions on the basis of our analysis of the literature and the variables in Appendix A. These are: (1) relation of the innovation to organizational domain, (2) integration, (3) risk, (4) need.3 These dimensions are explicated in the subsequent paragraphs. It is important to specify the relationship of the technological innovation to an organization’s domain. Domain refers to an organization’s sphere of activity-that is, the technologies employed, population served, and services rendered by an organization (Thompson, 1967). Warren elaborates that domain &dquo;includes the organization’s access to both input and output resources ... not only those resources needed for task perforbut also those needed for maintenance of the organimance zation itself’ (Warren, 1972: 22). Within this broad conceptual notion, our theoretical concern is to specify the context of use for the technology, given the organizational domain. In particular, the ...
465
technology might be perceived to serve task performance functions and/ or system maintenance functions, it might have either limited or pervasive impact on the organizational domain, and it might be used either more or less frequently. Adoption decisions are likely to vary with the particular relationship of each technological innovation to these aspects of organizational domain. The second conceptual dimension, integration, refers to factors that facilitate the discovery and implementation of an innovation by the organization. We expect that adoption is contingent upon not only an awareness of the innovation but also the ability to successfully internalize the innovation into organizational practice (Yin and Quick, 1977). Communication about the innovation might be facilitated by such factors as linkages into professional colleague networks and the proximity of suppliers of the technology. The capacity to successfully implement the innovation might be contingent upon the skill levels among current staff and upon the compatibility between the innovation and other technologies currently employed by the organization. The dimension of risk incorporates those factors which influence the perception that the innovation will produce expected results.4 Among the factors which might be considered are the financial costs of the innovation (relative to the organization’s total budget and relative to other possible innovations), the availability of slack resources, and the specificity with which the innovation can be evaluated. The assumption is that organizational decision makers will undertake, directly or implicitly, a calculation of the expected benefits and costs from adopting a new technology. The probability of adoption is expected to increase with decreases in the ambiguity of results and in the relative costliness of the innovation. The need for an innovation includes both objective factors and subjective assessments of the organizational requirements to be met by utilization of the innovation. In the absence of a profitability criterion for selection of public technologies, decision makers must rely on other indicators of the utility of an innovation. Need provides a substitute in the sense that it identi-
466
fies benefits that can be expected from the innovation, given its functional capabilities. Nelson and Winter (1975) have referred to these need factors as the &dquo;selection environment&dquo; for an innovation. Thus, the adoption of a computerized innovation will be more likely to the extent that it is responsive to the organization’s need to perform an information processing task with
greater rapidity, greater complexity, higher volume,
efficiency,
or
to the extent that managers are
or
greater
optimistic
about
its contributions. For the innovations in this analysis one might hypothesize various forms of the relationships between adoptability and each of the four dimensions. We agree with those who suspect that these relationships might be nonlinear functions (Downs and Mohr, 1976), but there is little existing research on public sector innovation to guide our formulation of the configurations that ought to be expected. Consequently, the broken lines in Figure 1 present our best estimation of these hypothesized relationships. While we predict a relatively linear relationship between adoptability and the relation of the innovation to organizational domain, the other three relationships are hypothesized to be nonlinear. For example, we predict that at some point adoptability will drop precipitously as the level of risk increases. We also hypothesize that adoptability, at its extremes, will be less responsive to changes in the integration dimension. Finally, we expect that the influence of need upon adoptability will be more substantial when need is high (which will increase adoptability) than when it is low. DEVELOPMENT OF THE INDEPENDENT VARIABLES
Three steps were taken to measure the four dimensions we have specified. First, independent variables relevant to the innovations in this study were identified and operationalized (see Appendix A). Second, these independent variables were classified into the domain, integration, risk, or need categories. Third, the variables in each group were factor-analyzed in order to reduce the variables into a smaller number of derived concepts.
467
Figure
1:
Hypothesized and Operational Relations Between the Four Dimensions and Adoptability
~.
Conceptual
r
Although the use of factor analysis is to classify variables requires the exercise of judgment (concerning the fit between the variable and the more abstract concept), this procedure has important advantages. It provides for a substantial reduction of the original data and at the same time assists in the creation of aggregated although still meaningful constructs based on the original variables. Moreover, it is possible to treat the factors themselves as independent variables at an intermediate level of abstraction between the four conceptual dimensions and the concrete indicators, and thus to assess their capacity to account for variations in
adoptability.
468
Factor analyses of the variable groupings resulted in eight factors: task-maintenance orientation of the innovation, professional infrastructure, staff competence, visibility of the innovation, tructure, staff competence, visibility of the innovation, supplier proximity, external funding, agency dominance, and uncertainty. No factor was derived from the need variables, and therefore the two single indicators were retained for statistical analysis. The factor loadings for the first three sets of variables are displayed in Appendix B. Since labeling and interpreting factor dimensions is judgmental, we briefly discuss below our interpretation of each of the factors.5 y~A’-~~~a~c~ orientation or~/a~b~ of //~ in/~One One factor, factor, termed task-maintenance 0/’ the novation, was derived from the four domain variables. This factor characterizes the application along a continuum from intensive use in agency task-related activities to occasional use in organizationwide maintenance activities. A maintenanceoriented application is exemplified by less frequent generation of data and use of outputs, by use to service &dquo;house-keeping&dquo; needs, and by a more pervasive scope of impact on the organization. In contrast, applications oriented primarily toward task accomplishment are used more directly in service delivery, generate and require outputs that are used more frequently, and have a direct impact on a smaller part of the organization. Four factors, which we term professional infrastructure, staff competence, visibility, and supplier proximity, were derived from the integration variables. The first factor, professional infrastructure, is defined by the loadings of URISA membership and URISA conference attendance, and it reflects variations in what Rogers and Shoemaker (1971) call &dquo;communication integration&dquo; and &dquo;cosmopoliteness.&dquo; Governments which score higher on this factor are seen to have more extensive channels for interpersonal communication with professionalized reference groups characterized by interest in automated applications. Staff competence, the second factor, reflects the technical competence attributed to the organization’s data processing staff, and it is determined by the loadings of staff experience and staff skill level. Third, the factor termed the visibility of an
469
the amount of attention an application in is reflected the loadings of professional communiattracts) cation and of the relative earliness of an application’s first adoption. Finally, the loadings on the measures of local computer mainframe suppliers and service/ sales offices on the fourth factor represent the proximity of suppliers. Three factors were derived from the risk variables: external funding, agency dominance, and uncertainty. On the first factor, the high loadings of the absolute and relative measures of external funding for electronic data processing (EDP) represent variations in the provision of financial resources for computing from external sources. The level of external funding may influence the organization’s capacity for, and thus the attractiveness of, the investment of resources in new computer applications. Second, the loadings of absolute and relative expenditure measures for the particular agency which might adopt the application is termed agency dominance. On this factor, the relative status of an agency is inferred from its resource allocation share within the local government. Third, uncertainty is the name given the concept derived from the negative loading of cost relative to other agency applications and from the positive loading of specificity of evaluation. The loadings indicate that both the relative magnitude of investment in an automated application and a knowledge of how explicitly the results can be evaluated will affect the degree of certainty that the desired results will be obtained by adopting the application.
application (that is,
.> .
.
’
RESULTS z
Our basic objective is to assess the patterns of relationships between the adoption of computer applications as innovations and the various explanatory variables. To accomplish this, it is useful to characterize the direction and form of these relationships and also to identify interactions among the explanatory variables. The individual components of the integration, risk, and need dimensions (that is, staff competence, professional
470
infrastructure, and so on) have been normalized and summed so that these components as well as the conceptual dimensions can
explanatory variables. Initially, the shape of the between adoptability and the explanatory variables relationships is specified graphically. In the second section, regression analysis is employed to estimate the amount of variance in adoptability that can be accounted for by the explanatory variables. Finally, interactive effects among the four conceptual dimensions are evaluated by using controls that stratify the values on each be treated
as
dimension. THE SHAPE OF THE ADOPTABILITY RELATIONSHIPS
First, the shape of the relationships between adoptability and the component factors of the integration, risk, and need dimensions are investigated. (The domain dimension, which is composed of only one factor, is discussed, along with the other three dimensions, in the following section.) Graphic representations of these configurations were developed, using crosstabulations between the dichotomized dependent variable and categorized values on each of the independent variables. These configurations are presented in Figure 2. The area under each curve represents the probability of adoption. Adoptability is not clearly linked with the extensiveness of the professional infrastructure, a variable which could indicate the government’s involvement with a social interaction network that might diffuse information about the innovation (Figure 2a). There are, however, limited relationships between a higher level of adoptability and both increased staff competence and greater proximity of suppliers of the technology. The most critical relationship among the components of integration is the substantial increase in innovation adoptability when the innovation’s level of visibility is higher. This linkage suggests that an innovation has a higher probability of adoption where it has enjoyed greater attention in professional media and where it was introduced relatively early. Clearly, these circumstances should make the innovation more visible and comprehensible to potential adopters.
471
Figure 2: Operational Relationships Between Adoptability and ponents of Integration (I), Risk (R), and Need (N)
the Individual Com-
There is a clear positive relationship between adoptability and the risk component of external funding (Figure 2e). Consistent with many other studies of innovation and with the studies of computer innovation, the presence of slack resources provided by an outside funding agency facilitates the adoption of the innovative device. Although agency dominance has no clear relationship to adoptability, the risk component we have termed uncertainty has a most interesting configuration. From the
472
mid-low levels to the high levels of the uncertainty variable, the relationship of uncertainty to adoptability is fully consistent with the reasonable expectation-the probability of adoption decreases with increased uncertainty about the cost-benefit ratio and the likely success of the innovation. However, adoptability also drops sharply at the lowest levels of uncertainty. That is, where the cost of the automated application (relative to the cost of others) is low and where quite specific evaluation of the application is possible, the probability of adoption is reduced. There is no obvious explanations for this unexpected finding regarding adoptability and low uncertainty. Perhaps there is a type of trivializing-the-innovation rationale among adopters which leads to the conclusion that if the magnitude of costs and benefits is small, then the innovation is not worth the effort. A more plausible explanation is that the very fact that specific evaluation of the innovation can be undertaken is viewed by potential adopters as a negative factor. If the potential adopter believes that a clear assessment of shortcomings in production efficiency could be undertaken, the innovation might seem less desirable since a &dquo;failure&dquo; would be difficult to obfuscate (Feller, 1977). Whatever the underlying dynamic, the shape of the relationship at the low end of uncertainty is somewhat surprising. The perceived effectiveness of the user represents a subjective appraisal, by the chief executive, of a specific agency’s production efficiency in the use of automated applications. The relationship in Figure 2i reveals that further adoptions are most likely in those agencies that are judged to have been most effective in applying computer technology in the past. A record of effectiveness might heighten the agency’s perception that it needs further automated applications, and/ or it might dispose top managers to support the manager’s expanded use of automation, viewing it as costeffective and need-driven. The objective measure of need for each specific automated application reveals that, across the entire set of innovations, increased need is related to increased adoptability only at the lower levels of need.
473
Integration
Dimension
configurations of the adoptability-conceptual dimension operational relationships are displayed in Figure 1. It is clear that only the curve for the integration dimension closely corresponds to our expectations. Except at the lowest and highest levels of integration, adoptability increases as a relatively linear function of greater integration. Adoption decisions do seem to be responsive to the basic capacity of the organization to discover, comprehend, and implement the automated application. The
Organizational
Domain Dimension
The operational relationship between adoptability and the innovation’s relation to organizational domain in Figure I is particularly intriguing. Rather than the expected linear relationship, the actual configuration is roughly an inverted U. This curvilinear relationship suggests that innovations which relate to either purely task aspects or purely maintenance aspects of an organization’s domain are less likely to be adopted than those innovations which serve both task and maintenance functions. The task end of this dimension, it should be recalled, involves applications which impact primarily upon the particular agency, which do not disturb organizational arrangements, and which respond to frequently recurring information processing needs of the unit. From a bureaucratic self-interest perspective, this is the type of innovation that would, ceteris paribus, be most attractive to the agency. But the scarcity of EDP resources in most local governments and the value attributed by resource allocators to &dquo;organizationwide&dquo; efficiencies often result in a bias toward developing applications that benefit a wider spectrum of government actors. Given these intraorganizational realities, it is plausible that the most generalized support and the highest priorities would tend to gravitate toward those applications which appear to serve a mixture of maintenance and task-oriented functions. And this is what Figure 1 a suggests. While some highly task-oriented applications have a reasonable probability of
474
it is the middle ranges of this dimension where the incidence of adoption is highest.
adoption,
Risk Dimension
The form of the risk relationship bears only partial similarity to the relationship hypothesized in Figure 1. Through the middle range of risk, adoptability does decrease with increasing risk, although not as precipitously as anticipated. At high levels of risk, however, adoptability remains relatively constant with increasing risk. These data are consistent with Feller’s (1977) suggestion that innovators in public agencies are not particularly risk averse, since institutional characteristics in the governmental context minimize the chance that innovation &dquo;failures&dquo; will be discovered and/ or that those responsible will be identified and sanctioned. However, at lower levels of risk, adoptability actually decreases with decreasing risk. The shape of the relationship with low risk seems to be substantially influenced by the unexpected association between adoptability and uncertainty discussed above. Thus, the pattern in Figure I reinforces the inference that low-risk innovations tend to be viewed either as ones likely to generate substantial benefits or as ones with potentially high costs to the adopting agency because impacts can be evaluated so unambiguously that the identification and attribution of failure is probable. Need Dimension
relationship in Figure 1, between need and adoptability, roughly similar to the pattern hypothesized. Adoptability, a positive function of need at the lower range of the need values. But it levels off and then has only a slightly positive The fourth
is is
across most of the need dimension. This suggests that, lower range of need, other factors are more critical the beyond than the level of need in determining whether a government will adopt a particular innovation application. While the agency is likely to ground its advocacy for an automated application
association
475
in factors of objective need and production efficiency, the development and implementation of new applications does not seem particularly contingent upon increased need across most of this dimension. STATISTICAL ASSESSMENT OF THE RELATIONSHIPS
and significance of the associations between adoptability and the variables in Figures I and 2 can be further assessed by the use of statistical techniques. Initially, zero-order correlations between adoption and each of the independent variables were computed. Of the variables in Figure 2, all were statistically significant (at the .001I level) except professional The
strength
infrastructure and agency dominance. The graphic analyses in Figure 2 suggested that some of the bivariate relationships might be nonlinear. Consequently, certain transformations were attempted for the variables of domain, visibility, uncertainty, and need.6 An exponential transformation did not markedly improve the relationship of visibility with adoptability, and a transformation of the need variable improved its correlation with adoptability slightly (from +.06 to +.09). A transformation of the domain variable altered its correlation with adoptability from .03 to -.07. And there was a quite substantial improvement in the correlation between adoptability and the transformation of the uncertainty variable. This association increased from -.13 for the original values to -.32 for a transformation accounting for the curve at the low uncertainty level. Clearly, it is difficult to specify an optimal transformation through inspections; but these results reinforce the inference from the graphic analyses that some significant determinants of adoptability are nonlinear. To evaluate the independent contributions of the individual component variables of the integration, risk, and need dimensions, multiple regression techniques are useful.7 Individual regression analyses for the component variables of each dimension, taking adoption as the dependent variable, revealed that integration is the most powerful explanatory dimension, given its
476
explain 15% of the variance in adoptability (as by the R2 statistic). Among the component factors of integration, visibility of the innovation is the most significant
capacity
to
measured
Consistent with earlier inferences, there is a considerable increase in the probability of adoption for an innovation which has been the subject of more extensive professional communication and which has had an earlier introduction. The risk dimension explains 10% of the variation in adoptability and is primarily a function of the transformed uncertainty factor, whose quite significant association with adoptability was noted above. The components of the need dimension have minimal explanatory power regarding adoptability, accounting for only 1 % of the variance. Utilizing all the significant component variables in a single regression analysis revealed that there are substantial, independent contributions to adoptability from the variables identified earlier as important.8 The probability of innovation adoption tends to increase where: ( 1 ) the innovation is more visible; (2) there is less uncertainty regarding its evaluation and cost; (3) the innovation is more maintenance-oriented than task-oriented; (4) there is greater objective need (in a production efficiency sense) for the innovation; and (5) suppliers of computer services are more proximate to the adopting government. The optimal linear combination of these variables, along with several others (that were not statistically significant), explained 25% of the variance in adoptability. Thus, a parsimonious set of variables does account for a considerable amount of the variation across a quite diverse range of computer applications. And if there are contingent relationships among the explanatory variables, it is likely that a linear model underestimates the variance explained by these four dimensions. Given this possibility, it is useful to explore the conditional relationships among the four conceptual dimensions. one.
477 ASSESSING INTERACTIONS AMONG THE FOUR DIMENSIONS
To test for interactions among the four dimensions in their associations with the adoptability of innovations, each dimension was stratified into high and low categories. The criterion for inclusion in either category was to include only those values more than one-half standard deviation above or below the mean. That is, the data in each part of Table I are based on only those cases which met this criterion on the relevant stratifying variable. The table reports the fraction of explained variance (the R2) in adoptability accounted for by each of the other three dimensions, stratifying on the fourth. In interpreting these relationships, we can explore the extent to which interaction exists between any two of the conceptual dimensions, but we cannot infer causality in the relationship between the dimensions.
Organizational
Domain
When the organizational domain variable is used to stratify the cases, there are several striking results. Table I reveals that the relationship between adoptability and the need dimension is not contingent upon the domain dimension; but the effect of organizational domain on the explanatory power of the other two dimensions is substantial. The integration dimension is able to explain about 50% more variance when the task-maintenance orientation values are low than when they are high. And a quite dramatic difference occurs with the risk dimension, where the Rz2 for adoptability and risk is only +.02 for low values of the domain dimension but is +.44 for high values. These results suggest that when an application relates to task accomplishment aspects of an organization’s domain (that is, when the innovation primarily serves a recurrent task activity within a single agency), factors relating to the visibility and successful implementation of the innovation are particularly important. But the level of risk and, we infer, the magnitude of benefits relative to costs and the probability of failure have no consistent linear effect on decisions regarding such task-oriented
478
TABLE 1
R2 s for Three of the Dimensions when Stratified Scores of the Fourth Dimension
by
These findings are consistent with Feller’s contention that consideration of the cost-benefit shortcomings of an innovation is not necessarily a major constraint when the innovation is internal to the agency. Since the innovation tends to impact only the specific agency (in our analysis, a task-oriented application), the agency is able to cover the failure of an innovation. Feller argues that for such an innovation, failure is most likely to be defined in terms of failure to implement successfully, rather than shortcomings in eventual benefits. Thus, the motivation to adopt the task-oriented innovation relates more to implementation issues (integration) than to impact con-
applications.
siderations
(risk).
When one examines the more maintenance-oriented innovations as opposed to the task-oriented ones, risk becomes the
479
critical dimension. While the implementation considerations inferred from the integration dimension have some influence, the concern with magnitude of costs relative to benefits and with uncertainty of impacts is central. One explanation of the centrality of risk to maintenance-oriented applications pivots upon the composition of &dquo;interested and attentive parties&dquo; to such an innovation. When the application is maintenance-oriented, it will have direct or indirect impacts on a wider spectrum of actors both within and external to the agency primarily responsible for its utilization. For example, an automated current balance report system implemented by the finance department will generate financial information which is distributed to all operating agencies. Thus there are multiple actors who attend to the impacts of the innovation as it is routinized and who are capable of evaluating the substance and quality of its impacts. Given the existence of multiple actors who can assess impacts, the adopting agency is likely to be considerably more concerned that the inadequate performance of an innovation could be discovered and communicated. In this case, those who are most accountable for the innovation’s impacts (and who are typically the pivotal adovcates for adoption) will tend to be much more responsive to considerations of risk.
Integration When cases are stratified on the basis of high and low scores on the integration dimension, other types of interactions are evident. The relationship between adoptability and the level of integration does not seem to be contingent upon either the organizational domain or the need dimensions. But risk accounts for a substantially greater proportion of the variance in adoptability when integration is high rather than low. Higher integration implies that there is more visibility, comprehension, and staff competency regarding the implementation and use of the innovation. In such cases, it should be possible to make a more accurate assessment of the implications of risk and also to react more &dquo;rationally,&dquo; or at least more sensibly, to those implications.
480
integration is low, there is limited understanding and insight about the innovation, and adoption decisions are likely to be more strongly influenced by criteria or attitudes that ignore or misperceive the factors of risk. Thus, high integration is likely to complement the importance of risk, but low integration might limit its salience to adoption decisions. But where
technical
Risk
The risk dimension has limited interactive effects with two of the three conceptual dimensions in their relationship with adoptability. The need dimension is not systematically associated with adoptability at either the high or the low levels of risk. And the integration dimension seems to be an important explanator of adoptability regardless of whether risk is high or low. However, stratifying the cases by risk underlines the interaction effects between risk and the relation to organizational domain. Under conditions of high risk, the domain orientation of the automated application accounts for very little variance in adoptability. That is, neither task- nor maintenance-oriented innovations are more consistently deterred when risk and uncertainty about impacts are high. But when risk is low, relation to domain accounts for 15% of the variance. Specifically, the probability of adoption of the more maintenance-oriented innovations increases if they are low risk. This is futher evidence that, as the impacts of an innovation are more pervasive and there are more interested and attentive parties, adopters are most favorably disposed toward those innovations characterized by a minimum of risk to the adopting unit. Need
When the cases are stratified by need, there is no alteration in the relationship between adoptability and organizational domain ; but the integration and risk variables are affected by the level of need. The integration dimension explains twice the variance in adoptability when need is low than it does when need
481
is high. It seems that the innovation’s visibility, staff competence, and supplier proximity are likely to create some pressures and/ or inducements for adoption even when there is little apparent need (as measured by a specific production-efficiency indicator). In contrast, at high levels of need the decision to adopt is less contingent upon the level of integration in the organization. There is also slight evidence that when need is greater, there is more willingness to take larger risks in the attempt to realize the potential benefits of the innovation. High need moderates the constraint of greater risk, and the level of integration becomes a more important factor when need is low. Finally, need has no systematic influence on the adoptability-organizational domain linkage. As in the early analyses, Table I provides little evidence that innovation decisions are particularly contingent upon need.
CONCLUSIONS
In the life history of an innovative within an organization, the adoption decision is only the beginning. But it is the sine qua non of all innovative activity, and thus an adequate understanding of the adaptability issue is an important research concern. This article has been an empirical analysis of the determinants of the adoptability of a range of technological innovations. The concept of adoptability refers to the probability that a particular innovation will be adopted by an organization, and the primary concern is to specify those attributes of the innovation or of the organization which seem to affect that probability. The technological innovations that have been examined in this article are an array of ten specific computer applications used by American local governments on a variety of functions. These automated applications have ranged from automated alias name files in police departments to applications which perform cash flow analyses for financial managers. The methodological framework has been grounded in recent refinements in innovation research, especially with respect to the measurement of attributes which are contingent upon the relationship between the particular innovation and the particular organizational context.
482
findings in this analysis have been quite intriguing. Initially, four broad conceptual dimensions which might explain variations in the adoptability of innovations were identified and operationalized. These dimensions reflected the relation of the innovation to the organization’s domain, the capacity of the organization to integrate the innovation, the level of risk attached to the innovation, and the organization’s need for the innovative. Given these dimensions, a large set of relevant explanatory variables were reduced into a smaller set of derived factors that characterized aspects of each dimension. Of the eight At
a
substantive level, many
individual factors and two variables of need which were then created, most (seven of the ten) had associations with innovation adoptability that were statistically significant. Further analysis, which regressed all these factors in a single analysis, revealed that four were particularly important. The data revealed that higher probability of adoption for a particular innovative computer application is associated with: ( 1 ) greater visibility of the innovation; (2) less uncertainty about the cost and evaluation of the innovation; (3) greater EDP staff competence to implement the innovation; and (4) a higher level of objective need for the innovation. When a combination of graphic and statistical analyses were undertaken on the four conceptual dimensions, interesting differences were revealed. The data revealed that innovations adoptability increases as a function of greater organizational integration. Of the four dimensions, the integration dimension has the highest independent explanatory power for variation in adoptability of the innovations, and it is also the most obviously linear of the four dimensions. Moreover, if one examines the components (factors) of integration, it is evident that an innovation’s adoptability is particularly influenced by its visibility and by the level of EDP staff competence. These findings are fully consistent with the multiple regression noted above. With respect to the dynamics underlying the process, it seems that the greater visibility of the innovation both increases the attention local actors accord to the innovation and enhances the adopter’s sense that they understand its implications for their government’s
483
among EDP staff also seems to increase the level of local awareness about the innovation, and it raises the level of conviction among local actors that the innovation can be successfully developed and implemented. The risk dimension also displays considerable explanatory power for variation in adoptability among innovations in their particular organizational contexts. This dimension’s effect corresponds broadly to the intuitive notion that as risk is greaterthat is, as there is more uncertainty about an innovation’s cost and benefits and as there is greater dependence on internal funding to support the innovation’s development-the probability of adoption tends to diminish. However, a more surprising finding is that at the lower levels of risk there is also a reduction in adoptability. We offered several plausible explanations for this finding, primarily relating to the view that the risk constraint might operate in peculiar ways to public organizations. This view suggests that decisions which are low-risk but are easily evaluated by external actors are somewhat unattractive to public agencies. These intriguing relationships between innovative and risk, which are explored further in the next paragraph, clearly merit additional research. It is also evident that adoptability varies substantially with respect to the organizational domain dimension. The graphic representation suggests that an innovation is most adoptable when it satisfies a mixture of task and maintenance functions. Such an application is desirable to the individual unit most directly served, since both advocacy of and responsibility for the innovation can be spread, and yet the unit remains the primary beneficiary of the innovation. And this kind of adoption is likely to be supported by other units which enjoy at least indirect benefits and by central decision makers who might see the adoption as serving broader organizational interests. It is the more pervasive, maintenanceoriented applications that are most powerfully influenced by the level of risk, since multiple actors are in a position to evaluate the effectiveness of the innovation and of the adopting agency’s utilization of it. While a more risky maintenance-oriented application is likely to find few promoters, a risky task-oriented
operations. Higher competence
484
innovation might be advocated by an agency which reckons that no outsiders will be capable of evaluating the agency’s success or failure. Indeed, among task-oriented applications, the least adoptable ones are those subject to the most specific and unambiguous impact eveluation. The fourth dimension, need, has the weakest associations with adoptability. Our specific measures of objective need for the
applications (relative to other local government’s needs for the same application) might indicate the production efficiency (or selection environment) for the innovation. The multiple regression equation reveals that greater need has a small but significant coefficient with respect to higher probability of adoption. But the broader inference from the data is that the need dimension is much less critical to variance in adoptability than are the domain, integration, and risk dimensions. From a public policy perspective, it seems that the adoptability of this class of innovations is responsive to some variables which can be manipulated by interventions. In particular, adoptability can be enhanced if high risk can be moderated, especially for high-need innovations. This might be accomplished by the provision of considerable external funding support for the innovation or by lowering the relative cost of the application in some other manner (such as the operation of an effective program for softwear transfer). Secondly, adoptability also seems to be quite affected by the visibility of the innovation. Thus, an innovation is more likely to be adopted to the extent that communication about it is facilitated within professional peer networks or that an efficient information &dquo;clearinghouse&dquo; for automated systems is instituted. Third, these findings suggest that the dynamics among key determinants of adoptability differ between task- and maintenance-oriented applications. Hence, attempts to manipulate the integration dimension for task-oriented applications and the risk dimension for maintenance-oriented applications have higher probability of stimulating adoption
decisions.
Finally, this analysis has noteworthy implications for conceptual and theoretical issues in innovation research. The re-
485
of specific independent variables into a of explanatory factors and four &dquo;superparsimonious group dimensions&dquo; has demonstrated one means to make analyses more manageable and yet retain meaningful concepts. It is significant that the relationships between adoptability and three of the four duction of
a
large
set
conceptual dimensions (relation to organizational domain, risk, need) were clearly nonlinear. Integration is the only explanatory dimension that had a relatively linear relationship with adoptability. These findings underline the methodological problems of empirical studies of innovation adoptability that rely upon statistical techniques measuring linear relationships. We had modest success in specifying the relationship between adoptability and a transformation of the risk variable; but the
and
effective use of variable transformations and nonlinear models in innovation research requires much attention. It is also clear from our analysis that the application of the types of interactive measures consistent with the innovative-decision design is feasible and that the design is a promising framework for developing more valid and generalizable findings across a class of innovative devices. Innovation research might become more coherent to the extent that these kinds of conceptual refinements are applied with consistency in forthcoming empirical studies of
adoptability.
NOTES
1. The concept of "information processing task" is an analytic taxonomy which is based on the primary modality of an automated application. It is discussed in Kraemer et al. (1976) and Danziger (1977). 2. The ten computer applications selected, indicative of six empirically derived diffusion patterns, were: (1) library periodical holdings, (2) data-processing data dictionary, (3) federal and state grant file(s), (4) alias name file, (5) cash management/cash flow
analysis, (6) wants/ warrants file, (7) employee records, (8) program structure related to line-item budget, (9) payroll preparation/accounting, and (10) real property records. Since the number of applications associated with each of the original categories of diffusion patterns were unequal, the selection procedure resulted in a disproportionate stratified sample. This procedure was used to avoid the type of selection bias identified by
486 Downs and Mohr ( 1976) and by Warner ( 1974). For a discussion of selection bias in innovation research, the development of the empirically derived diffusion categories and their description, see Perry and Kraemer (1978). 3. The choice of these dimensions is based on an analysis of the empirical studies of Bingham (1976) and Feller and Menzel (1976) and the research critiques of Warner (1974) and Downs and Mohr (1976). 4. See Fliegel and Kivlin (1966: 244). Our definition of risk differs from the classical economic use of risk in that it does not incorporate decision maker utilities or preferences. Rather, as it is used here, risk focuses on factors likely to influence decision maker certainty and concern about cause-effect relationships associated with the innovation’s costs and benefits. For a discussion of risk and innovation in state and local government, see Feller (1977). 5. Principal factoring with iterations was used to derive the factors in Appendix B. This method is based on the assumption of underlying commonalities among the variables. An orthogonal rotation in which estimates of commonalities were placed in the main diagonal of the correlation matrix was used to simplify the factor structure. All factors with an eigenvalue of less than I were deleted. The eigenvalue for each factor is presented in Appendix B. Separate factor analyses were performed for each group of variables associated with a particular conceptual dimension. 6. Such data transformations are discussed in Kruskal (1972: 182-192). 7. Multiple linear regression was used to explore the predictive power of the variables associated with the four conceptual dimensions. Although transformations of particular variables partially compensate for the linear relationships assumed by regression, they cannot overcome all the prediction problems associated with nonlinear relationships. Therefore, we do not report or discuss these regression results at length. Future studies may achieve better predictions by using more sophisticated nonlinear regression models. 8. The standardized beta coefficients and constants for this summary regression were: -.26 +.19 domain +.03 staff competence +.32 visibility +.05 supplier proximity +.02 external funding -.33 uncertainty + .07 need. The F-levels for the domain, visibility, uncertainty, and need factors were significant at p < .01. Supplier proximity was significant at p < .025.
REFERENCES BINGHAM,
R. D. (1976) MA: D. C. Heath.
Technological Innovation
in Local Government.
Lexington,
DANZIGER, J. N. ( 1977) "Computers, local governments, and the litany to EDP." Public Administration Rev. 37 (January/February): 28-37.
DANZIGER, J. N. and W. H. DUTTON (1977) "Technological innovation in local government: the
case
of
computing." Policy
and Politics 6
(September):
127-149.
DOWNS, G. W. (1976) Bureaucracy, Innovation and Public Policy. Lexington, MA: D. C. Heath.
DOWNS, G. W. and L. B. MOH R (1976) "Conceptual issues in the study of innovation." Admin. Sci.
Q.
21
(December): 700-714.
487
FELLER, I. ( 1977) "Uncertainty and risk in the adoption of technological innovations by state and municipal governments." Institute for Research on Human Resources, University Park, Pennsylvania. (mimco) FELLER, I. and D. C. MENZEL (1976) Diffusion of Innovations in Municipal Governments. Institute for Research on Human Resources, University Park, Pennsylvania. (1975) "Diffusion milieux as a focus of research on innovation in the public sector." Presented at the annual meeting of the American Political Science Association, San Francisco, September 2-5. FLIEGEL, F. and J. KIVLIN (1966) "Attributes of innovations as factors in diffusion." ---
of Sociology 82 (November): 235-248. Planning for Innovation Through Dissemination and Utilization of Knowledge. Ann Arbor: Univ. of Michigan, Institute for Social Research. HICKSON, D. H., D. S. PUGH, and D. C. PHEYSEY (1969) "Operations technology and organization structure: an empirical reappraisal." Admin. Sci. Q. 14 (September): Amer. J.
HAVELOCK, R. (1969)
378-397.
KRAEMER, K. L., J. N. DANZIGER, W. H. DUTTON, A. MOOD, and R. KLING ( 1976) "A future cities survey research design for policy analysis." Socio-Economic Planning Sciences 10 (September): 199-211. KRUSKAL, J. B. (1972) "Statistical analysis: transformations of data," pp. 182-192 in International Encyclopedia of the Social Sciences. New York: Macmillan and Free Press.
MANSFIELD, E. (1971) Technological Change. New York: Norton. MOHR, L. B. (1969) "Determinants of innovation in organizations." Amer. Pol. Sci. Rev. 63 (March): 111-126. NELSON, R. and S. WINTER (1975) "Growth theory from an evolutionary perspective: the differential productivity puzzle." Amer. Econ. Rev. 65 (May): 338-344. PERRY, J. L. and K. L. KRAEMER (1978) "Innovation attributes, policy interventions, and the diffusion of computer applications among local governments." Policy Sciences 9
(April):
179-205.
ROGERS, E. M. and F. F. SHOEMAKER (1971) Communication of Innovations. New York: Free Press.
THOM PSON, J. D. ( 1967) Organizations in Action. New York: McGraw-Hill. UTTERBACK, J. M. (1974) "Innovation in industry and the diffusion of technology." Science 183 (February): 620-626. WARNER, K. E. (1974) "The need for some innovative concepts of innovation: an examination of research on the diffusion of innovations." Policy Sciences 5: 433-451. WARREN, R. (1972) "The concerting of decisions as a variable in organizational interM. Tuite et al. (eds.) Interorganizational Decision Making. AVC. YIN, R. and S. K. QUICK (1977) "Routinization I: thinking about routinization." Rand Corporation, Santa Monica, California. (mimeo) YIN, R., K. HEALD, M. VOGEL, P. FLEISCHAUER, and B. VLADEK (1976) A Review of Case Studies of Technological Innovations in State and Local Services. Washington, DC: Rand Corporation.
action," pp. 20-32 in
Chicago:
488
489
490
491
492
Perry is Asso(-iate Professor of Administration in the Graduate School of Administration at the University of Cal(fornia, Irvine. He has authored and coawhored articles appearing in American Political- Science Review, Industrial Relations, Policy Sciences, Western Political Quarterly, and Public Administration Review. The research in he is currelllly engaged includes studies OrJ»1101’at1017 in local goB’emment and of llle impact 01’labor-i?7a?iageitieiii relationships on g01’ernl77elTlal efficiency and C’/~/~e(’lIl~PI7PS.s.. James L
James N. Danziger is Associate Professor o/’ Political Science in the School of Social Sciences at the University of California, Irvine. His research on local goaernment decision-making is i17lernatiullal in scope, and his rurrem research interests are focused on the impacts of computer technology on decision-making and UI1 operation in loc-al got,erpilyieiii. He is the author o/~ Making Budgets: Public Resource Allocation and has published articles in books and professional juurnals on the suhjec’ts ol’biit4getar.i, decision-making, policy anal_n.sis, and the use and impacts of computer technulogy.
