YOU ARE DOWNLOADING DOCUMENT

Please tick the box to continue:

Transcript
Page 1: 5866 1

COVER SHEET

Davidsson, Per (2005) Methodological Approaches to Entrepreneurship: Past and Suggestions for the Future. Small Enterprise Research 13:pp. 1-21. Copyright 2005 Per Davidsson Accessed from http://eprints.qut.edu.au

Page 2: 5866 1

Davidsson, P. 2005. Methodological Approaches to Entrepreneurship: Past and Suggestions for the Future. Small Enterprise Research 13, 1-21

Methodological Approaches to Entrepreneurship:

Past and Suggestions for the Future1

Per Davidsson Brisbane Graduate School of Business (BGSB) at Queenland University of Technology (QUT),

and the Jönköping International Business School, Sweden

1 This manuscript builds heavily on ideas previously developed in Davidsson (2003; 2004) and shares parts of the text with these works.

Page 3: 5866 1

Introduction This journal aims at developing and sharing some ideas on methodological development in entrepreneurship research, based on the author’s experience from two decades of empirical work in this area. As there is no shortage of ideas in the literature as regards what “entrepreneurship” is, and what entrepreneurship research should study, it may be useful to first set the stage on these issues. In order to facilitate cumulative knowledge building and enhance the legitimacy of entrepreneurship as a scholarly domain it is important that the phenomenon called "entrepreneurship" be defined as unambiguously as possible. The particular view embraced by the author regards entrepreneurship as "the creation of new [to the market] economic activity". This includes launching new products or services, new business models, radically changed price/value relationships, or entering the market as new competitor with essentially imitative products or services. The effects of such actions are that:

1. Customers get new choice alternatives, giving them more value for their money.

2. Incumbent firms get reason to in turn improve what they offer the market. 3. If successful, imitating followers will enter the market, further enhancing

the previous two effects. The end result is more efficient and/or effective resource utilization, i.e., increased wealth. As a consequence of adopting this perspective the author prefers to reserve the term "entrepreneurship" for market contexts, or at least market-like contexts. Entrepreneurial processes and behaviour share many of the characteristics of creative behaviour and processes in other domains, but in the absence of customers, competitors and potential followers (or close equivalents), entrepreneurship as portrayed above looses its precise meaning if these also be included.

This does not mean that entrepreneurship research should refrain from learning from creative behaviour and processes in, e.g., the arts, sports, politics, or even crime. It is just not necessary to use the label "entrepreneurship" for all manifestations of creativity or change. Neither does the above position mean that entrepreneurship research should confine itself to studying successful cases (that achieve the three effects above). In order to understand success (and the truth of what on the surface looks like 'failure') it is necessary to study unsuccessful cases as well. Therefore it is useful to distinguish between the definition of the phenomenon (new economic activities that lead to the one or more of the effects above) and the delineation of the scholarly domain. As to the latter, the below is suggested (cf. Davidsson, 2003; 2004):

Starting from assumptions of uncertainty and heterogeneity, the domain of entrepreneurship research encompasses the study of processes of (real or induced, and completed as well as terminated) emergence of new business ventures, across organizational contexts. This entails the study of the origin and characteristics of venture ideas as well as their contextual fit; of behaviors in the interrelated processes of discovery and exploitation of such ideas,

Page 4: 5866 1

and of how the ideas and behaviors link to different types of direct and indirect antecedents and outcomes on different levels of analysis. This domain delineation emphasizes the process character of entrepre-

neurship and allows for failure as well as studies on different levels of analysis (which correspond to knowledge interests in different disciplines). It also deliberately avoids associating entrepreneurship with a specific organizational context. However, the start-up of new, independent businesses remains a particularly central phenomenon in entrepreneurship research and accordingly an independent business context is explicitly or implicitly assumed in much of the remainder of this paper.

After this introduction the role of different basic approaches, e.g., “qualitative” and “quantitative” research will be discussed. A separate sub-section is devoted to laboratory research methods. The exposition then turns to method implications of accepting the domain delineation just presented. Separate subsections are devoted to general design, sampling, operationalization and analysis techniques.

“Qualitative” and “Quantitative” studies The Need for “Qualitative” Entrepreneurship Research

Arguably, knowledge development processes are incomplete without theory

testing. Nonetheless, both “qualitative” and “quantitative” research is helpful for gaining insight into entrepreneurship. Our total knowledge development requires the combination of different types of information. Researchers who say or think “I cannot see any meaningful knowledge coming out of that research approach” should realize that this may reflect a shortcoming of “I” and not necessarily of “that approach”.

There are some characteristics of the entrepreneurship research domain that point at a need for “qualitative” research. One is the relative youth of the field. We have simply not had time enough yet to familiarize ourselves with all facets of this empirical phenomenon, or to develop all the theory we need in an exploratory manner. Another is the heterogeneity of the phenomenon. If we only did research at arms-length distance there are the risks that because the relationships are different for different parts of the heterogeneous population the results would be either weak or true on average but not for most individual cases. Close-up information may be needed in order to learn about the heterogeneity, so as to make valid abstractions and generalizations. Further, entrepreneurship research often concerns events that are infrequent, unanticipated and/or extraordinary. Phenomena of this kind may be difficult to capture with conventional, “quantitative” approaches (cf. Baumol, 1983; Brymer, 1998). It is worth pondering that at the extreme of conventionalism, the most spectacular instances of entrepreneurship would invariably end up as disturbing and possibly deleted outliers in regression analyses (cf. the ‘Analysis Method’ section).

The process character of entrepreneurship may also call for “qualitative” approaches (cf. Brundin, 2002). For example, virtuous or vicious circles of events would be very hard to capture in “quantitative” work and entirely impossible with a cross-sectional survey design (Davidsson, 1986). Moreover, for matters that are in

Page 5: 5866 1

principle possible to capture in broadly based studies insights from cases are often an indispensable input to good design and interpretations. However, “qualitative” studies based on retrospective interviews share with cross-sectional surveys the severe limitations of selection and hindsight biases. Hence, longitudinal case studies of an ethnographic kind would seem a stronger methodological candidate for making progress in entrepreneurship research. Such studies can capture processes and involve direct and rich observation of real behavior. Hence, there are frequent calls (but little following; Aldrich & Baker, 1997) for intense ethnographic study of (super-) entrepreneurs of the kind (Mintzberg, 1974) did in his classical study of managers. Of course, like all in-depth case approaches such a study would share the limitation that statistical generalizability cannot be obtained. What can potentially be achieved in “qualitative” work is analytical generalizability – the generation of new concepts and suggested contingencies that are worthy of consideration also for cases not investigated. When entrepreneurship is regarded as equal to starting and running an independent business there is little doubt that ethnographic close-up studies of business owner-managers are an excellent tool for generation of new, tentative insights. When entrepreneurship is regarded as the process of creation of new economic activity the value of this approach is, regrettably, more limited. This is because “entrepreneur” is a transitory role; nobody is an “entrepreneur” all the time. Therefore, even if focused on an independent businessperson known for repeated success at creating new ventures there is considerable risk that intense, week-long observation of that individual would capture many more managerial than entrepreneurial behaviors. Unless the researcher is extremely persistent or lucky, s/he is likely to capture but a fraction of the entrepreneurial process. “Quantitative” vs. “Qualitative”—a Confused Debate What entrepreneurship research does not need is the often confused and confusing debate about qualitative versus quantitative research that goes on in business studies. The debate is here called confusing and confused firstly because it rarely recognizes that “quantitative” (and therefore “qualitative”) has several distinct meanings, and secondly because it often and non-justifiably equates the nature of the data with issues of philosophy of science, rigor, and depth.

Critics who favor qualitative approaches rarely distinguish between three distinct meanings of “quantitative”: using many cases (census studies or large samples); applying formal measurement (coding data in numerical form), and the use of statistical or mathematical analysis techniques. These different aspects of “quantitative” carry distinct advantages and shortcomings. In short, with many cases one gains generalizability and loses detail. With formal measurement and statistical techniques one gains a higher degree of objectivity and makes it possible to detect patterns that are otherwise beyond human cognitive abilities. On the other hand, the scope of the research is reduced to what is possible to measure or estimate with currently available techniques.

Further, the three aspects of “quantitative” do not necessarily go together. When a study is quantitative in the “many cases” sense it is for practical reasons often quantitative also in the second and third senses. Analyzing non-quantified data on several variables from hundreds of cases is beyond the limits of most researchers. It is

Page 6: 5866 1

not the case, however, that the use of formal measurement requires that many cases be analyzed. Case studies may be highly quantified in this sense, and here already the categories “quantitative” and “qualitative” get seriously blurred—which is why these terms are put within quotation marks. Within a case study there may also be room for application of statistical or other quantitative analysis methods, such as time series regression or techniques like multi-dimensional scaling or conjoint analysis, which allow analysis of single-respondent data (Hair, Anderson, Tatham & Black, 1998). Moreover, the development of formal techniques for recording and analyzing data that have traditionally been regarded “qualitative” makes the distinction along this dimension seem inadequate and obsolete (Miles & Huberman, 1994).

The second, confused and confusing part of the argumentation is the frequent assumption of a strong or even deterministic association between type of data and philosophy (or ideology) of science. Here “quantitative” is held to be almost equal to “positivistic” and “superficial” while “qualitative” data are associated with hermeneutics, phenomenology, or social constructionism, and with depth of analysis. Other parts of the research community would typically equate “quantitative” with “rigorous”, and consequently regard “qualitative” work as lacking rigor. It is admittedly likely that a true positivist researcher—if you could find one—would favor a quantitative approach in one or more of the above senses. It is further possible that an orientation towards phenomenology or social constructionism makes the researcher more inclined to use qualitative approaches, although it is difficult to see a binding logical connection. Studying a large number of cases does not seem to come naturally for a researcher with a hermeneutic bend. However, hermeneutics does not seem fundamentally or logically (albeit perhaps by convention) opposed to the use of numerically coded data or statistical analyses as steps on the way towards holistic interpretation. Over all, the choice of philosophical vantage point seems to have some, but far from completely deterministic, implications for the choice between “qualitative” and “quantitative”.

In the other direction there is no determinism whatsoever. Data do not know how they are going to be used! “Quantitative” data—i.e., measured and/or coming from many cases—do not make the researcher a positivist or the research deductive. There is nothing in the nature of the data that prevents the analyst from speculations about how the data or results generated inductively should be interpreted. A bad example is numerology; a good one is creative interpretation of patterns in the data that could not have been detected had not many cases been studied and sophisticated, computer-aided, multivariate analysis techniques been applied. The fact is that published “quantitative” research is full of exploratory findings and the use of techniques—e.g., cluster analysis and exploratory factor analysis—that a true positivist would deem unscientific. Moreover, it is perfectly possible for research involving many cases and/or formal measurement and/or statistical techniques to be sloppy rather than rigorous.

Likewise, the mere use of qualitative data does not make a researcher worthy of any honorary philosophical title. Irrespective of the type of data, the researcher may confess to any philosophical congregation—or be an onto- and epistemological orphan or bastard. Again, the data don’t know what we are going to do with them. As regards rigor, what many (European, and possibly Australian) researchers regard as a bias—on the part of US-based journals, for example—against “qualitative” work

Page 7: 5866 1

should arguably instead be interpreted as skepticism against research that does not show enough rigor. With rigor is then meant well-founded rationales for the selection of cases; systematic and transparent procedures for data collection; application of formal techniques for analysis of the data, and the like. It is perfectly possible to apply rigor in a study using data from few cases; non-numerical data, and without applying statistical methods for data analysis. Contrary to popular belief, “qualitative” research that is both rigorous and deep is appreciated by large parts of the entrepreneurship research community, including journal reviewers and editors. On the other hand, it is perfectly possible to be superficial with “qualitative” data. There appears to be some confusion between “depth” and “detail” on this issue. “Qualitative” data typically include more detail. Depth is another thing, and mainly an outcome of the time, effort and talent that is put into the data collection, analysis and interpretation work. There is no inherent property in “quantitative” data that prevents one from going deep in the analysis and interpretation. Bad Research Practice: Addressing “Quantitative” Questions with “Qualitative” Research While entrepreneurship research arguably needs both “qualitative” and “quantitative” approaches, there has to be a proper match between the research question and the chosen approach. The main problem this author sees with “qualitative” research as actually practiced is that researchers often apply such approaches to make claims about issues their approach is fundamentally inadequate for addressing. A little anecdote about a presentation of a study of business founders at a conference a couple of years ago can illustrate this. The cases were chosen because the founders were female and the start-ups were in a particular, recently deregulated industry. The data were collected through retrospective interviews. Several of the interviewees reported they had difficulties obtaining the bank loans they needed, and when prompted some of them ascribed this to the fact that they were women. Because of this, the researcher publicly claimed that women entrepreneurs were discriminated against by the banks.

Now, saying that banks systematically discriminate against a particular group is a very serious accusation, and because people have a high degree of faith in what researchers say, researchers should be careful not to make strong claims like this on the basis of very shaky—or in this case no—evidence. If the goal is to establish whether women business founders have difficulty obtaining bank loans because they are women, then for a minimum one needs to a) investigate a group of subjects that is representative for the category “women business founders”, b) measure the frequency of loan refusals, and c) compare the results with a group of men business founders. There can be absolutely no escape from these requirements. In addition, one should preferably also be able to rule out other substantive explanations or that the group difference could easily be due to stochastic variation. In this case all that the researcher had were a few women from a judgment (or convenience) sample saying they had problems getting loans, and feeling this might have something to do with the fact that they were women.

While there is solid research evidence elsewhere that women are discriminated against in society in other ways, and while this in conjunction with previous private experiences by these women may have made their suspicion of discrimination a

Page 8: 5866 1

reasonable hypothesis, the above is not the quality of evidence needed for researchers to make strong claims about sex discrimination. To make matters worse, the accusation made by the researcher was probably a false one in this case, as comprehensive and systematic research on precisely that matter and from the same country was published at about the same time, arriving at the conclusion that women entrepreneurs were not discriminated against by banks (Björnsson, 2001). A review of the international entrepreneurship research literature on the issue of gender discrimination by banks suggest that the support is very limited, and when gender differences appear they do in most cases not seem to be attributable to gender per se (Ahl, 2002).

More generally, research questions that are inherently quantitative in nature need quantitative research to be answered. Questions about quantitative differences (more; better; stronger; more often, etc.) between groups, or about such within-group changes over time, are inherently quantitative in nature. So are questions about the form, direction and strength of relationships between variables. In order to make claims about such, variables need to be measured and the relationships among them estimated with some kind of analysis technique. When a researcher makes claims like “x has a strong, positive effect on y” (e.g. “the entrepreneurs’ persuasive skills have a profound effect on their level of success”) on the basis of research where no formal measurement or estimation was involved, it just exemplifies application of extremely crude and unreliable measurement and estimation instruments, namely the researcher’s subjective assessments.

This is not to say that qualitative research has no role in a problem like potential gender discrimination by banks. A piece of useful research that would be classified as “qualitative” by conventional criteria and which would get at issues that are unlikely to be within reach for a survey approach would be a participant observation study, where the loan officers’ way of talking to and about male and female loan applicants were studied. If there were discrimination, such a study would not only give strong indications of this fact, but also offer an opportunity to understand (and/or generate concepts and hypotheses about) the mechanisms behind it, which is precisely what qualitative research can excel at. In order to impress a researcher of this author’s ilk, however, the presented evidence should not just be a number of illustrative quotes that support the researcher’s hypothesis, but convincing evidence that the loan officers’ treatment of women applicants was systematically different, and that this was to their disadvantage.

Arguably, the most fruitful way forward for entrepreneurship research for the future would be integrated research programs that included several types of research addressing different aspects of the same issues. This would make for real cross-fertilization between different approaches, rather than having different camps of researchers develop separate discourses that are ignored by the other camps. Because of this author’s own balance of expertise and ignorance, respectively, most of the remainder of this paper will deal with method issues pertaining to “quantitative” research. This is not due to lack of appreciation of qualitative work. On the contrary, favorite references include for example Bhave (1994), McGrath (1999), Sarasvathy (2001), Shane (2000) and Van de Ven, Polley, Garud & Venkataraman (1999. It is for people with other expertise this authors’, however, to suggest future developments based on qualitative work specifically.

Page 9: 5866 1

Laboratory Research Methods Experimental and simulation-based research was rare in the early days of empirical entrepreneurship research, but has recently become more frequent (Baron & Brush, 1999; Fiet & Migliore, 2001; Gustafsson, 2004; Sarasvathy, 1999a). An important reason to welcome this development is that studies of real-world entrepreneurs or intrapreneurs in their real settings do not necessarily provide leads to normatively valid knowledge about entrepreneurship. They may be wrong in what they do, or at least sub-optimal. Additional reasons for experimental approaches include the general strength of such methods in establishing causality, and the process nature of entrepreneurship. Studying real world processes is a costly and time-consuming endeavor with uncertain rewards. Although the laboratory alternatives will never completely substitute for real world studies they are a valuable complement—and an acceptable alternative when resource limitations prohibit a longitudinal study in the real setting. Laboratory methods make it possible to compress time and collect multi-period data without having to wait for ages before any serious analysis work can be done. However, it would be a mistake to believe that laboratory research is easy work. Laboratory research tends to be heavy in the front end, in the design of the simulation or experiment, whereas actual data collection and analysis can be less of a burden relative to other methods.

The suggested domain delineation portrays entrepreneurship as consisting of two interrelated and overlapping sub-processes, discovery (idea development) and exploitation (making it happen). Both of these could presumably be induced in the laboratory. However, the laboratory alternative may be particularly suitable for the earliest phases of the discovery process. Ideas for new ventures do not occur very frequently and are also for other reasons difficult to capture with real time approaches (cf. Simon in Sarasvathy, 1999b, p. 52). Therefore, laboratory research may be better suited to cover this part of the entrepreneurship research agenda.

The general shortcoming of laboratory research is that the external validity of the findings can always be questioned. What works in the laboratory does not necessarily repeat itself in the field, where many other influences determine outcomes. Therefore, laboratory work should preferably be integrated into programs that include also analysis of real world data, so that the field and the laboratory can inform and inspire one another. Interesting ideas along these can be found in Cialdini’s (1980) reasoning on “full cycle social psychology”. Entrepreneurship Research as the Study of Processes of Emergence of New Ventures: General Design Issues What are the methods consequences of a research focus as implied by the domain delineation suggested in the introduction? The keyword new ventures suggests that in order to belong in the entrepreneurship domain, the research has to give explicit consideration of new venturing. As long as this requirement is fulfilled, the research can be conducted on any level of analysis—individual, firm, industry, region, nation, or something else (cf. Davidsson & Wiklund, 2001). That is, the research design should at least include the middle box in Figure 1. Preferably, the research should pay

Page 10: 5866 1

attention to antecedents and outcomes as well, but this is not indispensable in the same way. On the individual level, the research thus cannot be confined to, for example, owner-managers’ personal characteristics on dimensions assumed to be entrepreneurial, as related to the size (an outcome) of their businesses. In order to qualify as entrepreneurship research there should be assessment of the middle box, i.e., new venturing activities by these individuals. On the region level, studying the relationship between structural characteristics of regions and their economic growth (or well-being) does not become entrepreneurship research until the quantity and/or quality of new economic activity on in the regions is introduced as the mechanism of such a relationship.

The emphasis on processes implies that longitudinal research is needed. Such

studies have not been common in entrepreneurship research (Aldrich & Baker, 1997; Chandler & Lyon, 2001). In order to establish causality the alleged cause must precede the ensuing effect. In cross-sectional research it is usually unknown whether this truly is the case. To take an entrepreneurship example of this problem, consider the hypothesis in early entrepreneurship researcher that entrepreneurs were characterized by a more internal Locus-of-Control (Brockhaus, 1982). Individuals with more internal locus-of-control have stronger belief in their own ability to control their, as opposed to it being directed by fate or powerful others. Some cross-sectional studies have supported the idea that business founders and/or owner-managers have a more internal orientation than others. The problem is that such an orientation would be a likely outcome of being a business owner-manager, as opposed to being a subordinate within a hierarchy. Hence, a positive correlation is not enough. In the absence of longitudinal research, showing that internal locus-of-control precedes business founding the belief that an internal orientation causes individuals’ choices of an entrepreneurial career remains a hypothesis.

Further need for longitudinal design stems from the fact that the study of processes involves more than static comparison of a beginning state and an end state. Many things happen between the initiation of a venture start-up process and its completion or termination (Bhave, 1994; Carter et al, 1996; Davidsson & Honig, 2003; Davidsson & Klofsten, 2003; Delmar & Shane, 2002, 2004; Gartner & Carter, 2003; Katz & Gartner, 1988; Sarasvathy, 2001; Van de Ven et al, 1999). Therefore, longitudinal designs with repeated assessment of the ventures’ development over time are needed in order to adequately capture those processes.

The emphasis on emergence suggests that new ventures should be studied from early on in the process (cf. Davidsson, 2003a, 2003b). How can one study emergence? Retrospective studies would be subject to severe selection and hindsight biases. It is therefore preferable to study the processes as they happen, or as close to that ideal as possible. Regarding hindsight bias, it is well known in cognitive psychology that

Characteristics of entity X

Emergence of new ventures within or associated with entity X

Figure 1 Entrepreneurship research design possibilities

Outcomes on different levels

Page 11: 5866 1

memory is constructive in nature (Anderson, 1990). This means that no matter how honest and careful respondents are, they will still distort the image of what happened during the start-up process. Dead ends and detours will likely be forgotten and certain actions will be ascribed a rationale that only emerged afterwards. Such problems can to some extent be remedied through triangulation (second informant; written documentation), but in retrospective designs serious distortions are likely to remain regardless of such efforts. Selection bias concerns the need to study also “unsuccessful” or prematurely terminated processes. For one thing, this is needed in order to acknowledge the fundamental uncertainty that we highlighted in the domain delineation. If only completed start-up processes were studied it might be forgotten that completion is by no means a certain outcome for the newly initiated project.

The problem of selection bias is potentially even more serious than hindsight bias. In order to illustrate this, consider the following example. Imagine a research team that wanted to study “factors that lead to success in gambling”. The specific empirical context chosen is the horse race track. They design the study so that only those gamblers who actually won are included, and thus left the day at the races with a net gain (cf. only those founders who actually got their venture up and running). Analyzing our data, they arrive at the following conclusions:

a. Betting on horses is profitable b. The more you bet, the more you will win c. The more unlikely (higher odds) winners you bet on, the more you will

win

While true for winners, these conclusions are, of course, blatantly false inferences for the entire population of gamblers. On average, gamblers do not win. The more they bet the more they lose, and the proportion of gamblers who loses is larger among those who bet on long shots. But since we study only winners, Points a, b and c are the results that will be generated. The fact is that by studying only those processes that led to a successful start-up the researcher takes the risk of producing equally invalid results.

It can be argued that the need for longitudinal, concurrent data is relatively more pronounced for micro-level studies than for more aggregate levels of analysis. For studies of the latter kind, a methodology like that employed by the Global Entrepreneurship Monitor (GEM) may suffice (Reynolds et al, 2001). That is, a design that aims at cross-sectional comparison of the prevalence of a) on-going start-up processes and b) recently completed start-ups across countries, industries, regions, or perhaps even such a disaggregate level as firms.

An interesting but unusual type of study uses the venture idea itself as the unit of analysis. That is, “entity X” in Figure 1 is the new venture idea and the activity and new organization that evolves around it. For such studies, which would follow new, emerging business activities from their conception and through whatever changes in human champions and organizational home that might occur, the importance of early catch and multi-wave concurrent data collection seem particularly important. This is also a type of study where “qualitative” work in the form of longitudinal case studies is needed, not least for aiding with the difficult problems of design and interpretation of “quantitative” studies with this focus.

Page 12: 5866 1

Sampling Issues Social Science Is Not Opinion Polls In quantitative research, representative sampling and associated significance testing are important safeguards against ignoring relevant parts of empirical populations, giving undue weight to atypical cases, or ascribing substantive meaning to results that can easily have been generated by chance factors. However, for the statistical inference apparatus to be applicable in a strict sense the population should be well defined and the sample should reflect the composition of this population in a probabilistically known manner. These are ideals that are rarely achieved in social science research, and even less in entrepreneurship research. Response rates in the typical 5-35 percent range are already reason to view application of statistical inference as highly dubious. To make matters worse, statistical inference theory is a tool that is tailor made for opinion polls and industrial quality control rather than for the true needs of a social science researcher (cf. Cohen, 1994; Oakes, 1986). In a political opinion poll the population is clearly defined and reachable: all eligible voters. What the poll is after is their political preferences on the day of investigation. Hence, applying statistical inference theory, the uncertainty of results for a random sample is known, and hence it is possible to determine whether or not a difference between two political parties, or the change for one party over time, deserves a substantial interpretation. Probability sampling and significance testing are useful tools in this situation. Much more can be said on basis of this probability sample than on the basis of just any equally sized voter sample of unknown origin.

There are occasions in entrepreneurship research that are very similar to this situation. The country comparisons of the prevalence of “nascent entrepreneurs” in GEM are an example of this (Reynolds et al, 2001). Here, the interest is in what proportions of the adult population in various countries are involved in business start-ups at a given point in time. For the most part, however, entrepreneurship research like all social science research is not like opinion polls, and theories are not built by democratic vote. One implication of heterogeneity it that is not a given that every empirical case should be deemed equally important for theory building and theory testing. Social science research aims for theoretical representativeness—that the studied cases are relevant for the theory about to be tested or developed. There is no way a random sample can be drawn directly from the theoretical population, because that population does not exist in one place at one time. This is why, arguably, every empirical population, even if investigated it in its entirety, is a non-random sample from the theoretically relevant population. Therefore, the inference to the theoretical population is always an analytical and judgmental one, and not one based directly on the rules of statistical inference.

To illustrate the limitations of simple random sampling, consider a random sample of small firms, here meaning commercially active firms with less than 50 employees. Were such a sample drawn in Sweden it would have the following composition of firm sizes: 62 percent self-employed without employees; just short of 35 percent micro-firms with 1-9 employees, and a remainder of less than 4 percent

Page 13: 5866 1

firms with 10-49 employees (NUTEK, 2002). There is every reason to ask whether the solo self-employed are economically and theoretically sixteen times more important just because they are sixteen times as many as the “large” small firms. For most conceivable research questions they are not. As regards regions, if a country consists of ten large and economically growing regions and 90 small and backward ones, should all regions weigh equally in the analysis? If the size differences should be considered, by what criterion should the regions be weighted in the analysis?

This line of reasoning leads to the conclusion that the more important issue about sampling is theoretical representativeness. Simple random sampling is not necessarily the ideal. Stratified and deliberately “narrow” statistical samples, and even judgment samples, may on theoretical grounds be preferable. That is, the researcher should be able to defend that the elements in the sample represent the type of phenomenon that the theory makes statements about. This is equally relevant for case study research. Moreover, replication—not statistical significance testing—is the crucial theory test. The development and testing of sound theory requires replication in several sub-groups of analyzable size within the same study, as well as across several studies that investigate theoretically relevant samples from different empirical populations. The importance of replication is further elaborated in Davidsson (2004; Ch. 9). Sampling Issues on Various Levels of Analysis A common approach in early entrepreneurship research, and still a common first backbone reaction among newcomers to the field, is to start from a sample of “entrepreneurs”—meaning people who are currently running their own business—with a group of “non-entrepreneurs”. The assumption is that characteristics of the person explain their behavior. The problem is that this approach confounds at least four factors: a) the innate propensity to engage in a business start-up; b) the ability to attain enough success for the venture to survive; c) the propensity to persist in the face of failure, and d) a range of situational factors that contribute to engaging, succeeding and persisting in entrepreneurship regardless of personal characteristics. In short, this type of contrast is not a very sound approach. More promising individual level approaches do not reduce entrepreneurship to a dichotomy but starts from any sample and aim at explaining individuals’ degree of entrepreneurial behavior, and their success at such. In practice, the sample would usually be one of individuals whose situational conditions give them discretion to act entrepreneurially, e.g., a sample of business owner-managers with varying degrees of innovativeness and success. If the approach is to compare categories, contrasts between habitual or expert vs. novice entrepreneurs (Gustafsson, 2004; Ucbasaran et al, 2001) appear more fruitful than are comparisons of entrepreneurs vs. non-entrepreneurs.

A particularly relevant and at the same time difficult research task is the sampling of emerging new ventures. On this level, the Panel Study of Entrepreneurial Dynamics (Gartner et al, 2004; Reynolds, 2000) has made a major contribution by developing a technique for capturing (reasonably) representative samples of on-going start-up efforts. This eliminates huge issues of selection and hindsight biases. The approach has recently been expanded to the study of new internal ventures (Chandler et al, 2003). An inescapable limitation of the approach is that as the nature of the

Page 14: 5866 1

ventures cannot be determined beforehand it is difficult to apply criteria based on theoretical relevance as discussed above.

Issues of relevance also apply when sampling firms, industries, regions and other aggregate units. For example, the sampling of firms often implies tricky matching problem between theoretical and empirical definitions of “firm”, which are commonly neglected (Davidsson & Wiklund, 2000). On the industry or regional levels the statistical organization’s grouping into categories may likewise not be the most relevant in the light of the specific research questions asked (Davidsson, Lindmark & Olofsson, 1994). Apart from relevance, questions concerning size, size distribution and other heterogeneity need to be considered. As regards size the sampled units have to be large enough to show a reliable level of entrepreneurial activity within the time frame studied. Regarding size distributions, a problem that has to be solved is what weight should be given in the analysis to the smallest and largest firms, industries and regions when the size variation can be a factor of ten or more. As regards other heterogeneity, the researcher has to consider whether variables like “growth in assets” or “number of new products launched” apply and have the same meaning, e.g., across firms in different industries.

To sum up, there is reason in future entrepreneurship research to pay more attention to the theoretical representativeness of the sample used. Further, entrepreneurship research can be studied on many levels of analysis, and the individual level approach to compare alleged “entrepreneurs” with “non-entrepreneurs” is probably one of the least fruitful. Recent developments have envisaged a way to sample and follow emerging business ventures, and this type of sampling arguably holds much promise. Operationalization Issues

The fundamental issue for this sub-section is that entrepreneurship research often includes complex and difficult-to-assess issues. Consequently, the use of simplistic single-item, ad hoc measures of focal variables belongs in history. In order to move the field forward, researchers need to show greater sophistication in measurement than was the case in the early phases of entrepreneurship research. Space limitations regrettably prohibit a complete treatment here. The discussion will be confined to some overarching balancing problems that researchers face when designing their studies. One such balancing problem is that of using a previously validated measure versus developing a new one.

Consider, for example, the items displayed in Table 1, which are four out of the nine items of the 1989 version of the frequently used Entrepreneurial Orientation (EO) scale (Wiklund, 1998). The EO measure can be accused of all sorts of shortcomings. The items displayed seem to be a mix of preferences, past behaviors and beliefs. Further, Lumpkin & Dess (1996, 1997) argue that item (f) gauges “competitive aggressiveness” rather than pro-activeness. This item as well as the first innovation item (a) also show poor technical properties as they do not load neatly on the intended factors in a factor analysis or contribute positively to Cronbach’s Alpha (Brown et al, 2001). Finally, while the measure is intended to be a firm level measure, usually only one person’s answers are used to represent the firm’s views.

Page 15: 5866 1

Table 1 Sample items from the Entrepreneurial Orientation scale Generally our company prefers to... a) Strongly emphasize the marketing of the company's present products.

1 2 3 4 5 6 7

Strongly emphasize R&D.

How many new kinds of products or services has your company introduced over the past 5 years? b) A lot of new products/services.

1 2 3 4 5 6 7

No new products/services.

f) Normally our company tries to avoid overt competition, but rather takes on a "live-and-let -live"-position.

1 2 3 4 5 6 7 Normally our company takes on a very competitive oriented "beat-the-competitor"-position.

Generally we believe that... h) The business environment of the company is such that fearless and powerful measures are needed to obtain the company's objectives.

1 2 3 4 5 6 7

The business environment of the company is such that it is better to explore it carefully and gradually in order to achieve the company's objectives.

For these reasons, it would be tempting for a researcher to develop a new and

better measure for assessing entrepreneurial tendencies on the firm level. However, the EO measure is not without theoretical backing (Miller & Friesen, 1978, 1982). Further, if a couple of items are dropped it has acceptable internal consistency both as a three-dimensional and as a one-dimensional construct (Brown et al, 2001). More importantly, it has been shown to have theoretically meaningful relationship in a range of previous studies (Wiklund, 1998). There exists quite a body of knowledge about the direct, moderating and mediating effects of EO in different settings, and rich possibilities for comparing new results with established ones.

With a new measure all these advantages would be lost. Moreover, developing useful new measures is harder than most researchers believe until they have tried. It is possible to develop alternatives to EO that have some advantages, but this is something that requires considerable effort (cf. Brown et al, 2001).

A partly related balancing exercise concerns the choice between the “perfect” operationalization for a specific type of venture, and the most generally applicable operationalization. This, again, is one of the method consequences of the heterogeneity of the entrepreneurship phenomenon, and it was briefly touched upon in the sampling section above. For example, the best measure of firm size may be the number of vehicles for a taxi company, the number of seats for a restaurant, and the quantity of electricity delivered for a power station. But how are we to compare the firms’ growth across these different measures? Sales and number of employees are more generally applicable, but may have other disadvantages (Bolton, 1971; Davidsson & Wiklund, 2000).

Consider also the list of indicators of entrepreneurial action in Table 2. Assume we took a simple summation of the number of yes responses as the measure of firm level entrepreneurship. Would this be an adequate or “fair” measure? Doing (b) makes sense as an expansion strategy only if the markets for inputs or outputs are such that local presence is necessary. How does a retailer respond to (g)? Would not larger

Page 16: 5866 1

firms be likely to tick more affirmative responses simply because they do more things due to their sheer size; not because they are somehow “more entrepreneurial”? There is little doubt that the scale would be sensitive to size and industry.

Table 2 Possible indicators of entrepreneurial activities

Have you during the last 12 months carried out any of the following measures? a. Started doing business with a country you previously had not done business with?

Yes No

b. Started operations in a new location in Sweden? Yes No

c. Started marketing yourselves in a new way? Yes No

d. Carried out a considerable change of the company's organization? Yes No

e. Carried out a considerable change in the company's internal operations? Yes No

f. Introduced an important new product or service or in any other way substantially changed what you offer the customers?

Yes No

g. Commenced development of a new important product, service or similar, which has not yet been introduced?

Yes No

h. Carried out measures in advance, which you think you would otherwise have been forced to do sooner or later?

Yes No

i. Carried out changes specifically in order to get ahead of competitors? Yes No

One would hope that including several alternative manifestations of

entrepreneurship reduces the industry biases, but a size effect is inescapable. In developing a scale of this kind, the less-than-perfect options available seem to be the following:

1. Develop one operationalization that is assumed to be good for all ventures/firms, and accept that interesting manifestations of entrepreneurship that clearly apply only to narrow subsets of firms cannot be included. Also accept that larger firms and firms in some industries, on average, exercise more entrepreneurship than do smaller firms and those in certain other industries.

2. Develop one operationalization for all ventures/firms. Normalize the score as deviations from industry/size class (or other) averages. This eliminates some bias, but at the cost of only allowing for within-group and not between-group differences.

3. Develop separate and adapted operationalizations for different sub-groups (by industry, size class, or otherwise). Express this measure in terms of standard deviations, so that comparisons can be made across different operationalizations of entrepreneurship. This allows including the most relevant indicators for each category, but involves considerable risk of comparing apples with oranges.

Page 17: 5866 1

Similar issues of best vs. most generally applicable operationalizations could be discussed with regard to explanatory or outcome variables. As no obviously “right” decision can be made on this kind of issue a good solution is often to try different approaches within the same study—if space allows.

In summary, early entrepreneurship research was plagued with a lack of established operationalizations. As a result weak, ad hoc measures yielded weak and conflicting results. Today’s researcher is in a more fortunate position. There is much more to build on, and the researcher should think at least twice before deciding to develop her own measures. There is still need and room for new, validated measures, though. In trying to develop new measures, researchers are encouraged to think far beyond their own current study and really make an effort to develop measures that are anchored in theory. They are also well advised to carefully consider across what parts of empirical populations the new measures are applicable. Analysis Method Issues Heterogeneity and Analysis Method The domain delineation proposed in the introduction suggests that heterogeneity be accepted as a fundamental vantage point in entrepreneurship research. This has implications not only for sampling strategy but also for analysis. For example, the fact that micro level studies almost never reach more than 50 percent explanation of variance is not only due to the usual suspects like model misspecification and measurement errors. It is reasonable to assume—unlike underlying assumptions of microeconomic theory and various statistical techniques—that the effect of variable x on variable y is different for different cases (cf. Katona, 1974, 1975). Hence, regression coefficients and the like normally represent average effects for members of the investigated population. Around this average there is considerable variation, leaving variance unexplained (cf. Davidsson, 1991). The reason why one variable seems relatively more important than another is not necessarily that for every respondent the first factor is more important than the second. It may just as well be the case that relatively more managers consider the first factor at all. For a distinct minority, the on average less important factor may be the determining variable.

Heterogeneity can be dealt with in the analysis in an array of ways. One is to accept it and its effect on explanatory power. Another it is to conduct sub-sample analyses in order to see what the relationships look like for different, more homogeneous, sub-groups (Brown et al, 2001; Davidsson, 1991). A particularly relevant example here is Samuelsson’s (2001; 2004) work, where he shows that innovative and reproducing venture ideas, respectively, follow venture creation processes that are substantially different.

Yet another way to deal with the heterogeneity of the effect of predictor variables, which has become rather common in recent years, is to explicitly model it in a moderated regression analysis (Brown, 1996). There are also more creative, non-standard approaches that can be used. A particularly exemplary demonstration is Gimeno’s et al (1997) careful adaptation of analysis tools to the analysis problem. In particular, their study is exceptional in its attention to heterogeneity regarding what is deemed an acceptable level of success.

Page 18: 5866 1

Analysis Implications of the Minority Nature of Entrepreneurship Entrepreneurship research often takes an interest in the exceptional. Often the analysis is aimed at learning about a small minority that stands out from a larger population: individuals currently involved in a start-up; venture-capital backed ventures; IPOs; rapidly growing firms, or some other select minority. Again, this has implications not only for sampling but also for analysis method.

For example, one major research question in the Panel Study of Entrepreneurial Dynamics (PSED; cf. Gartner et al, 2004) concerns who is more likely to be a nascent entrepreneur. This is a single digit minority of the general population. The typical analysis method for this type of research question, logistic regression analysis, does not provide unbiased estimates when the group sizes are this uneven. In addition, the optimal function may be one that reaches a high overall correct classification rate by putting almost all cases in the less entrepreneurial group, i.e., by performing poorly with respect to what was the researcher’s key interest. There often exist—albeit perhaps not in the standard statistical packages—specific techniques that can solve this problem. For example, Wagner (2004) recently applied Rare Events Logistic Regression (King & Zeng, 2001a, 2001b) in order to get unbiased estimates when analyzing German data from the Global Entrepreneurship Monitor (Reynolds et al, 2001).

Although the perspective on entrepreneurship advocated in this paper acknowledges also relatively mundane, imitative venture start-ups, it is no doubt the case that the most interesting (and infrequent) instances of entrepreneurship are found at the other outskirt of distributions. This insight reveals a very fundamental problem with the conventional statistical analysis techniques that most researchers master. These variance-explaining techniques typically focus on central tendencies, preferably for normally distributed variables. Outliers are technical problems to be eliminated; not thrilling empirical phenomena of the highest societal import. In sharp contrast, the key interest in an entrepreneurship study may well rest with the rare cases at the high end of a highly skewed distribution. Thus, to make relevant contributions in future research, investigators may have to make some investment in learning new analysis techniques. There are alternatives to opening the tin can with a hammer.

Analysis Implications of Entrepreneurship as Process

The domain delineation in the introduction also portrays entrepreneurship as a process. This, too, has consequences for choice of analysis techniques. While it is possible to attain new insights about entrepreneurial processes by applying conventional techniques to process data (Carter et al, 1996; Davidsson & Honig, 2003), the way forward is to apply techniques that can make full make full use of the longitudinal aspects of the data. Two (sets of) promising techniques in this context that are Event History Analysis (Blossfeld & Rohwer, 2002) and Longitudinal Growth Modeling (Muthén & Curran, 1997; Muthén & Khoo, 1999).

In Event History Analysis the data set is organized as monthly spells (or other periodicity). The technique makes use of the longitudinal aspect of the dependent as well as independent variables. Independent variables can be entered as time invariant

Page 19: 5866 1

or time variant. In the latter case the value of the independent variable is allowed to change over time. The dependent variable changes its value in the period when the event to be predicted has occurred. Cases where the event has still not occurred when the last data collection is made are treated as right censored – a problem the technique is designed to deal with. Previously often applied to medical treatment (of potentially lethal conditions) the logic of the technique makes it especially suited for predicting abandonment (vs. continuation) of the start-up processes. However, it can also be applied for analyzing, e.g., ‘up and running’ vs. ‘still trying’. See Delmar & Shane (2002, 2003a, 2003b, 2004) for relevant applications.

While the dependent variable in Event History Analysis is always dichotomous, the independent variables can be either dichotomous or continuous. The technique can be regarded a longitudinal alternative to logistic regression. When the dependent variable is continuous Longitudinal Growth Modeling (LGM) is a particularly interesting alternative. In the context of new venture emergence the dependent variable may be, for example, the accumulation of gestation activities in PSED-like research; the gradual attainment of the cornerstones of Klofsten’s Business Platform Model (cf. Davidsson & Klofsten, 2003), or any other variable that is analogous to growth. LGM shares some characteristics with regular structural equation modeling techniques like LISREL and thus has the advantages of being suitable for models with latent variables and indirect as well as direct relationships. The technique aims at predicting both initial situation and development over time. Therefore it can at least to some extent handle the problem that cases are at different stages of development when sampled. A shortcoming of LGM is that unlike Event History models it cannot include cases that dissolve during the studied period in the analysis. In order to avoid erroneous conclusions based on success bias the LGM analysis should therefore be supplemented with other types of analyses of the discontinued cases. This makes it possible to rule out that these share the same characteristics that appear as success factors in LGM. For a relevant application see Samuelsson (2001; 2004).

Finally, while partially missing data (internal non-response) is always a problem that has to be dealt with in data analysis, it is aggravated when the data are longitudinal. With multiple waves of data the likelihood that a case has complete information on every variable included in the analysis asymptotically approaches zero. The problem of attrition—that some cases are lost entirely over time—is worsened when loss of cases due to partially missing data are added. As a result there may be too few cases left for a meaningful analysis. This calls for methods that allow the inclusion of cases with partially missing information. Replacing missing data with the mean or with a predicted value from a regression reduces the error variance, and therefore it approaches fraud to apply such practices when data are partially missing for a large number of cases. Fortunately, method experts have developed more sophisticated techniques for data imputation that can be applied. For a more elaborate treatment see for example Little (1987), Fichman & Cummings (2003), or other contributions to the same special issue of the Organization Research Methods journal (Vol. 6; Iss. 3).

Page 20: 5866 1

SUMMARY AND CONCLUSION This paper has advocated the following delineation of the field of entrepreneurship research:

Starting from assumptions of uncertainty and heterogeneity, the domain in of entrepreneurship research encompasses the study of processes of (real or induced, and completed as well as terminated) emergence of new business ventures, across organizational contexts. This entails the study of the origin and characteristics of venture ideas as well as their contextual fit; of behaviors in the interrelated processes of discovery and exploitation of such ideas, and of how the ideas and behaviors link to different types of direct and indirect antecedents and outcomes on different levels of analysis.

It has further been argued that knowledge development in this domain benefits

from different types of research—“qualitative” as well as “quantitative”, and laboratory research as well as studies that rely on data from the real setting. Preferably, these different types of research should be combined in comprehensive research programs. It would at least be advantageous if the different forms of research informed and inspired one another, rather than different methodological camps or entrepreneurship researchers developing separate and non-communicating discourses.

It was also argued that entrepreneurship research can, and should be conducted on different levels of analysis. Regardless of the level chosen the study has to take new venturing on the studied level into explicit consideration in order to qualify as entrepreneurship research. The focal phenomenon should not be reduced to an assumption. On each level there are certain sampling and operationalization issues to deal with. While statistical sampling and testing are important tools in research they are not the ideal tools that solve every problem. When the two do not fully coincide, it was argued that theoretical relevance of the sample is more important than statistical representativeness.

Further, as entrepreneurship is a process, longitudinal design are required. Preferably the data should also be concurrent rather than retrospective, so as to emphasize the foci on emergence, uncertainty and outcome variability, and avoid biases stemming from hindsight and selection of successful cases only. Longitudinal, real time designs also call for new analysis methods. Event History Analysis and Longitudinal Growth Modeling were presented as promising alternatives for making fuller use of longitudinal data. The process nature of entrepreneurship also brought forth the need for data imputation techniques.

Regarding the heterogeneity of entrepreneurial phenomena it was argued that this can be dealt with through narrow sampling, sub-sample analysis, or by explicitly modeling differential influence of the predictor variables. The perhaps most resisted conclusion regarding analysis methods emanates from the discussion of the minority characteristic of entrepreneurship. This is that the standard set of variance-explaining techniques, which focuses on central tendencies, assumes normal distributions, and regards outliers a problem, may be fundamentally inadequate for many analysis tasks in entrepreneurship research. This does not mean that quantitative entrepreneurship research should be abandoned. It just means that the researcher’s toolbox may need some overhaul.

Page 21: 5866 1

References Ahl, H. J. (2002). The Making of the Female Entrepreneur. Doctoral dissertation,

Jönköping International Business School, Jönköping. Aldrich, H. & Baker, T. (1997). Blinded by the cites? Has there been progress in the

entrepreneurship field? In D. Sexton & R. Smilor (Eds.), Entrepreneurship 2000 (pp. 377-400). Chicago, IL: Upstart Publishing Company.

Anderson, J. R. (1990). Cognitive Psychology and its Implications. New York: Freeman.

Baron, R. A. & Brush, C. G. (1999). The role of social skills in entrepreneurs' success: evidence from videotapes of entrepreneurs' presentations. In P. D. Reynolds, W. D. Bygrave, N. M. Carter, S. Manigart, C. M. Mason, G. D. Meyer & K. G. Shaver (Eds.), Frontiers of Entrepreneurship 1999. Wellesley, MA: Babson College.

Baumol, W. J. (1983). Toward operational models of entrepreneurship. In J. Ronen (Ed.), Entrepreneurship (pp. 29-47). Lexington, MA: Lexington Books.

Bhave, M. P. (1994). A process model of entrepreneurial venture creation. Journal of Business Venturing, 9, 223-242.

Björnsson, B. (2001). En bank även för kvinnor? Småföretagares erfarenheter av rådgivningsmöten i bank (A Bank Also for Women? Small-Business Customers' Perceptions of Counseling Encounters in Banks). Doctoral dissertation, Göteborg University, Sweden.

Blossfeld, H.-P. & Rohwer, G. (2002). Techniques of event history modeling: new approaches to causal analysis. Mahwah, NJ: Lawrence Erlbaum Associates.

Bolton, J. E. (1971). Small Firms. Report of the Committee of Inquiry on Small Firms. London: Her Majesty's Stationery Office.

Brockhaus, R. H. (1982). The psychology of the entrepreneur. In C. A. Kent, D. L. Sexton & K. H. Vesper (Eds.), Encyclopedia of Entrepreneurship (pp. 39-71). Englewood Cliffs, NJ: Prentice Hall.

Brown, T. (1996). Resource Orientation, Entrepreneurial Orientation and Growth: How the Perception of Resource Availability Affects Small Firm Growth. Unpublished Doctoral dissertation, Rutgers University, Newark, NJ.

Brown, T., Davidsson, P. & Wiklund, J. (2001). An operationalization of Stevenson's conceptualization of entrepreneurship as opportunity-based firm behavior. Strategic Management Journal, 22(10), 953-968.

Brundin, E. (2002). Emotions in Motion. Leadership during Radical Change. Doctoral dissertation. Jönköping: Jönköping International Business School.

Brymer, R. (1998). Hanging out with the good 'ole boys, gangsters, and other disreputable characters: field research, quantitative research, and exceptional events. In S. Grills (Ed.), Doing Ethnographic Research: Fieldwork Settings (pp. 143-162). Thousand Oaks: Sage.

Carter, N. M., Gartner, W. B. & Reynolds, P. D. (1996). Exploring start-up event sequences. Journal of Business Venturing, 11, 151-166.

Chandler, G. N., Dahlqvist, J. & Davidsson, P. (2003). Opportunity recognition processes: A taxonomic classification and outcome implications, Academy of Management Meeting. Seattle.

Page 22: 5866 1

Chandler, G. N. & Lyon, D. W. (2001). Methodological issues in entrepreneurship research: the past decade. Entrepreneurship Theory & Practice, 25(4, Summer), 101-113.

Cialdini, R. B. (1980). Full cycle social psychology. In L. Beckman (Ed.), Applied Social Psychology Annual (Vol. 1). Beverly Hills, CA: Sage.

Cohen, J. (1994). The earth is round (p<.05). American Psychologist, 47(12), 997-1003.

Davidsson, P. (1986). Tillväxt i små företag: en pilotstudie om tillväxtvilja och tillväxtförutsättningar i små företag (Small Firm Growth: A Pilot Study on Growth Willingness and Opportunity for Growth in Small Firms) (Studies in Economic Psychology No. 120). Stockholm: Stockholm School of Economics.

Davidsson, P. (1991). Continued entrepreneurship: Ability, need, and opportunity as determinants of small firm growth. Journal of Business Venturing, 6, 405-429.

Davidsson, P. (2003a). The domain of entrepreneurship research: Some suggestions. In J. Katz & D. Shepherd (Eds.), Advances in Entrepreneurship, Firm Emergence and Growth (Vol. 6, pp. 315-372). Oxford, UK: Elsevier/JAI Press.

Davidsson, P. (2003b). Method issues in the study of new venture start-up processes. Paper presented at the NFF Conference, Reykjavik.

Davidsson, P. (2004). Researching Entrepreneurship. New York: Springer. Davidsson, P. & Honig, B. (2003). The role of social and human capital among

nascent entrepreneurs. Journal of Business Venturing, 18(3), 301-331. Davidsson, P. & Klofsten, M. (2003). The Business Platform: Developing an

instrument to gauge and assist the development of young firms. Journal of Small Business Management, 41(1), 1-26.

Davidsson, P., Lindmark, L. & Olofsson, C. (1994). Dynamiken i svenskt näringsliv (Business Dynamics in Sweden). Lund, Sweden: Studentlitteratur.

Davidsson, P. & Wiklund, J. (2000). Conceptual and empirical challenges in the study of firm growth. In D. Sexton & H. Landström (Eds.), The Blackwell Handbook of Entrepreneurship (pp. 26-44). Oxford, MA: Blackwell Business.

Delmar, F. & Shane, S. (2002). What founders do: A longitudinal study of the start-up process. In P. D. Reynolds et al (Ed.), Frontiers of Entrepreneurship Research 2002 (pp. 632-645). Wellesley, MA.

Delmar, F. & Shane, S. (2003a). Does business planning facilitate the development of new ventures? Strategic Management Journal, 24, 1165-1185.

Delmar, F. & Shane, S. (2003b). Does the order of organizing activities matter for new venture performance? In P. D. Reynolds et al (Ed.), Frontiers of Entrepreneurship 2003. Wellesley, MA.: Babson College.

Delmar, F. & Shane, S. (2004). Legitimating first: Organizing activities and the survival of new ventures. Journal of Business Venturing, 19, 385-410.

Fichman, M. & Cummings, J. N. (2003). Multiple imputation for missing data: Making the most of what you know. Organizational Reserach Methods, 6(3), 282-308.

Fiet, J. O. & Migliore, P. J. (2001). The testing of a model of entrepreneurial discovery by aspiring entrepreneurs. In W. D. Bygrave, E. Autio, C. G. Brush, P. Davidsson, P. G. Green, P. D. Reynolds & H. J. Sapienza (Eds.), Frontiers of Entrepreneurship Research 2001 (pp. 1-12). Wellesley, MA.

Page 23: 5866 1

Gartner, W. B. & Carter, N. (2003). Entrepreneurial behavior and firm organizing processes. In Z. J. Acs & D. B. Audretsch (Eds.), Handbook of Entrepreneurship Research. Dordrecht, NL: Kluwer.

Gartner, W. B., Shaver, K. G., Carter, N. M. & Reynolds, P. D. (2004). Handbook of Entrepreneurial Dynamics: The Process of Business Creation. Thousand Oaks, CA: Sage.

Gimeno, J., Folta, T. B., Cooper, A. C. & Woo, C. Y. (1997). Survival of the fittest? Entrepreneurial human capital and the persistence of underperforming firms. Administrative Science Quarterly, 42, 750-783.

Gustafsson, V. (2004). Entrepreneurial Decision-Making. Doctoral dissertation, Jönköping International Business School, Jönköping.

Hair, J. F., Anderson, R. E., Tatham, R. L. & Black, W. C. (1998). Multivariate Data Analysis (5 ed.). Upper Saddle River, NJ: Prentice Hall.

Katona, G. (1974). Psychology and consumer economics. Journal of Consumer Research, 1, 1-8.

Katona, G. (1975). Psychological Economics. New York: Elsevier. King, G. & Zeng, L. (2001a). Logistic regression in rare events data. Political

Analysis, 9(2), 137-163. King, G. & Zeng, L. (2001b). Explaining rare events in international relations.

International Organization, 55(3), 693-715. Little, R. J. A. & Rubin, D. B. (1987). Statistical Analysis with Missing Data. New

York: Wiley. Lumpkin, G. T. & Dess, G. G. (1996). Clarifying the entrepreneurial orientation

construct and linking it to performance. Academy of Management Review, 21(1), 135-172.

Lumpkin, G. T. & Dess, G. G. (1997). Proactiveness versus competitive aggressiveness: Teasing apart key dimension of an entrepreneurial orientation. In P. D. Reynolds, W. D. Bygrave, N. M. Carter, P. Davidsson, W. B. Gartner, C. M. Mason & P. P. McDougall (Eds.), Frontiers of Entrepreneurship Research 1997. Wellesley, MA: Babson College.

McGrath, R. G. (1999). Falling forward: Real options reasoning and entrepreneurial failure. Academy of Management Review, 24(1), 13-30.

Miles, M. B. & Huberman, A. M. (1994). Qualitative Data Analysis (2 ed.). Thousand Oakes: Sage.

Miller, D. & Friesen, P. H. (1978). Archetypes of strategy formulation. Management Science, 24(9), 921-933.

Miller, D. & Friesen, P. H. (1982). Innovation in conservative and entrepreneurial firms: two models of strategic momentum. Strategic Management Journal, 3, 1-25.

Mintzberg, H. (1974). The Nature of Managerial Work. New York: Harper & Row. Muthén, B. O. & Curran, P. J. (1997). General longitudinal modeling of individual

differences in experimental designs: A latent variable framework for analysis and power estimation. Psychological Methods, 2(4), 371-402.

Muthén, B. O. & Khoo, S.-T. (1999). Longitudinal studies of achievement growth using latent variable modeling. Learning and Individual Differences, 10, 73-101.

Page 24: 5866 1

NUTEK (2002). Företagens villkor och verklighet 2002. Dokumentation och svarsöversikt (Conditions and Reality of Small Firms 2002. Documentation and Overview of Responses). Stockholm: NUTEK.

Oakes, M. (1986). Statistical Inference: A Commentary for the Social and Behavioural Sciences. Chichester: Wiley.

Reynolds, P. D. (2000). National panel study of US business start-ups. Background and methodology. In J. A. Katz (Ed.), Advances in Entrepreneurship, Firm Emergence and Growth (Vol. 4, pp. 153-227). Stamford, CT: JAI Press.

Reynolds, P. D., Camp, S. M., Bygrave, W. D., Autio, E. & Hay, M. (2001). Global Entrepreneurship Monitor. 2001 Executive Report. Kansas MO.: Kauffman Foundation.

Samuelsson, M. (2001). Modelling the nascent venture opportunity exploitation process across time. In W. D. Bygrave, E. Autio, C. G. Brush, P. Davidsson, P. G. Green, P. D. Reynolds & H. J. Sapienza (Eds.), Frontiers of Entrepreneurship Research 2001 (pp. 66-79). Wellesley, MA.

Samuelsson, M. (2004). Creating New Ventures: A Longitudinal Investigation of the Nascent Venturing Process. Doctoral dissertation, Jönköping International Business School, Jönköping.

Sarasvathy, S. (1999a). Decision making in the absence of markets: An empirically grounded model of entrepreneurial expertise. School of Business: University of Washington.

Sarasvathy, S. (1999b). Seminar on research perspectives in entrepreneurship. Journal of Business Venturing, 15, 1-57.

Sarasvathy, S. (2001). Causation and effectuation: towards a theoretical shift from economic inevitability to entrepreneurial contingency. Academy of Management Review, 26(2), 243-288.

Shane, S. (2000). Prior knowledge and the discovery of entrepreneurial opportunities. Organization Science, 11(4), 448-469.

Shane, S. & Eckhardt, J. (2003). The individual-opportunity nexus. In Z. J. Acs & D. B. Audretsch (Eds.), Handbook of Entrepreneurship Research (pp. 161-194). Dordrecht, NL: Kluwer.

Shane, S. & Venkataraman, S. (2000). The promise of entrepreneurship as a field of research. Academy of Management Review, 25(1), 217-226.

Ucbasaran, D., Westhead, P. & Wright, M. (2001). The focus of entrepreneurship research: Contextual and process issues. Entrepreneurship Theory & Practice, 25(4, Summer), 57-80.

Wagner, J. (2004). Are Young and Small Firms Hothouses for Nascent Entrepreneurs? Bonn, Germany: Forschungsinstitut zur Zukunft der Arbeit.

Van de Ven, A. H., Polley, D., Garud, R. & Venkataraman, S. (1999). The Innovation Journey. Oxford: Oxford University Press.

Wiklund, J. (1998). Small firm growth and performance: Entrepreneurship and Beyond. Doctoral dissertation, Jönköping International Business School, Jönköping.


Related Documents