Small Business Economics

, Volume 33, Issue 2, pp 151–163

Screening item effects in estimating the prevalence of nascent entrepreneurs

Authors

    • Management and International BusinessFlorida International University
Article

DOI: 10.1007/s11187-008-9112-3

Cite this article as:
Reynolds, P.D. Small Bus Econ (2009) 33: 151. doi:10.1007/s11187-008-9112-3

Abstract

The use of human population surveys to estimate the prevalence of nascent entrepreneurs has become a major feature of both longitudinal studies of the firm creation process, such as the US Panel Studies of Entrepreneurial Dynamics (PSED) research program, as well as cross-national comparisons, as reflected in the Global Entrepreneurship Monitor (GEM) initiative. The basic procedure has been to use interview screening items to locate individuals that may be considered candidate nascent entrepreneurs; other criteria are then used to identify those considered active nascent entrepreneurs. In these human population surveys, little attention has been paid to the potential impact of variations in wording in the initial screening items, either across time in the same language or in different languages, on the final prevalence rates. Analysis of 134 independent samples in the US over the 1993–2006 period, where different screening items were employed, indicates a major impact of item wording. Once adjustments to account for item variation were made, there was no statistically significant change in the prevalence of active nascent entrepreneurs, from 5 to 6 per 100 over the 1998–2006 period. This pattern of temporal stability is consistent with three other national programs measuring U.S. new firm creation activity.

Keywords

EntrepreneurshipNew firms: startupsSurvey methodsInformal economyUnderground economyEconomic sociology

JEL Classifications

L26M13C42E26Z13

1 Introduction

The idea of a nascent entrepreneur, a person actively involved in the creation of a new firm, is useful for both practical and theoretical objectives (Gartner 1988). Despite exhortations that new firm creation is not a core feature of entrepreneurial phenomena (Shane and Venkataraman 2001), it has become a central focus of empirical research among entrepreneurial scholars (Davidsson 2004). Locating nascent entrepreneurs has become a critical objective of a number of longitudinal, or panel, studies of new firm creation, many based on the initial US project, the first Panel Study of Entrepreneurial Dynamics or PSED I (Gartner et al. 2004; Reynolds 2000; Reynolds et al. 2004b). The second US panel study (PSED II) was implemented in 2005 and is currently in process. In addition, cross-national comparisons of 55 countries that are the focus of the Global Entrepreneur Monitor (GEM) initiative1 utilize estimates of the prevalence of nascent entrepreneurs and new firms to create the Total Entrepreneurial Activity (TEA) index (Reynolds et al. 2004a).

The basic research procedure has been to use surveys of the human population to identify those that consider themselves involved in the creation of a new firm and then use precise criteria to identify those that might be considered active nascent entrepreneurs. Much attention has focused on these later criteria; little attention has been given to the potential impact of variation in the screening items. Data from five US programs over 13 years provide a large number of independent national samples, 134, using the same basic protocol and the full range of screening items.2 Analysis suggests that population prevalence estimates are affected by the wording of the screening items.

1.1 Finding nascent entrepreneurs

Operational procedures for locating nascent entrepreneurs have been evolving since the first efforts were tested in Wisconsin in 1993 (Reynolds and White 1997). Estimates of the prevalence of nascent entrepreneurs are based on surveys of the human population, sometimes called household surveys, and the procedures employed in interviews designed for typical adults. The basic strategy is to ask questions—screening items—that will identify the largest possible range of individuals as “candidate nascent entrepreneurs.” The second stage involves applying additional criteria to identify “active nascent entrepreneurs” from the pool of candidate nascent entrepreneurs (Curtin and Reynolds 2004; Reynolds and Curtin 2004). The final estimates of the prevalence rates—the number of active nascent entrepreneurs per 100 adults in the population—could be affected by both the screening items and the nature of the criteria used to select active nascent entrepreneurs. This discussion focuses on the impact of alternative versions of the initial screening items.

Over the duration of two complementary research programs, PSED and GEM, attempts to expand the range of individuals and the scope of activities eligible for inclusion as a nascent enterprise have resulted in adjustments in both the wording and number of screening items. To minimize costs, one screening item was initially used, focusing on personal efforts to pursue new firm creation, identified in data sets as BSTART. Shortly thereafter a second item was added, asking about efforts to start a new firm as part of a job assignment, labeled BJOBST. After it was discovered that many individuals who considered themselves as running a going business were really in the start-up phase, a third item was added to locate owner-managers, labeled OWNMGE.

There have been revisions in the wording of the screening items. For example, three versions of wording for the initial, BSTART, item have been used:
  1. A.

    Are you, alone or with others, now trying to start a new business?

     
  2. B.

    Are you, alone or with others, currently trying to start a new business, including any form of self-employment?

     
  3. C.

    Are you, alone or with others, currently trying to start a new business, including any form of self-employment or selling any goods or services to others?

     
There were, in addition, two versions of the second, BJOBST, item:
  1. D.

    Are you, alone or with others, now trying to start a new business or new venture for your employer? An effort that is part of your job assignment

     
  2. E.

    Are you, alone or with others, now trying to start a new business or a new venture for your employer, an effort that is part of your normal work?

     
One version of the third screening item, OWNMGE, has been used in these projects:
  1. F.

    Are you, alone or with others, currently the owner of a business you help manage, including self-employment or selling any goods or services to others?

     
One benefit of harmonized research programs is the potential for considering the changes in response patterns over time. If the item wording has had a major effect, it would be possible to observe shifts in patterns over short time intervals that may be attributed to changes in wording. As the prevalence rates tend to be rather low, 5–10 per 100 respondents, slight changes in wording could have a significant effect.
To explore this issue, data files from five different research programs, all using a common paradigm, were assembled to provide 134 independent samples of the US adult population, one was restricted to the state of Wisconsin, the other 133 of the contiguous 48 US states and the District of Columbia. One sample was about 750 respondents, all other samples of about 1,000. The various screening items as used in these different samples are presented in Table 1.
Table 1

US nascent screening samples and item wordings

Project

Number of samples

Time period

BSTART item

BJOBST item

OWNMGEItem

PSED: Early development (1)

1

1993

A

PSED: Early development

1

1993

A

PSED: Early development (2)

1

1996

A

D

PSED I

30

1998

A

D

PSED I

29

1999

A

D

PSED I

3

2000

A

D

GEM: US

1

1998

A

E

GEM: US

1

1999

A

E

F

GEM: US

2

2000

B

E

F

GEM: US

3

2001

B

E

F

GEM: US

7

2002

C

E

F

GEM: US

9

2003

C

E

F

US Assessment

14

2004

C

E

F

PSED II

24

2005

C

E

F

PSED II

7

2006

C

E

F

(1) Wisconsin only sample

(2) Sample of 750

The temporal pattern for the proportion of BSTART “yes” responses from 1993 through 2006 is provided in Fig. 1, the 95% confidence intervals indicate the extent to which there are statistically significant differences between quarters. The wording of the BSTART items, A, B, or C, is indicated by the final character in the column labels across the horizontal axis.
https://static-content.springer.com/image/art%3A10.1007%2Fs11187-008-9112-3/MediaObjects/11187_2008_9112_Fig1_HTML.gif
Fig. 1

Positive response to BSTART item: US 1993–2006

There would appear to be a gradual increase in the prevalence of YES responses from 1993 through the first quarter of 2000 with a major increase to the third quarter of the same year. This doubling of the proportion of yes responses, from 6 per 100 to 12 per 100, appeared to have occurred with a change in BSTART wording from the A to B form, when the “self-employment” phrase was added. There was a slight drop after the 1st quarter of 2001, when the BSTART wording was changed to the C format. There is a strong possibility that a change in wording of the BSTART item may have had a major impact on the proportion of yes responses.

The same analysis and procedure was used for the BJOBST item, and the patterns are presented for 1996 through 2006 in Fig. 2. This item, with very minor differences in wording, shows exactly the same temporal pattern. Setting aside the unusually high prevalence from the GEM 1999 first quarter survey, there is a major increase in “yes” responses occurring in 2000 and continuing thereafter. This increase is from about 4 per 100 to 6 per 100.
https://static-content.springer.com/image/art%3A10.1007%2Fs11187-008-9112-3/MediaObjects/11187_2008_9112_Fig2_HTML.gif
Fig. 2

Positive response to BJOBST item: US 1996–2006

1.2 Analysis of wording effects and passage of time

Because the adjustment of the wording of the screening items to expand the scope of eligible activity occurred over time, it is possible that an increase in the prevalence of yes responses may reflect changes in item wording or an actual increase in entrepreneurial activity. To sort out the relative impact of these two factors, regression analyses were completed with all available samples for the three different items. Assuming that items C, E, and F were the baseline for item wording, dummy variables were included to represent the use of other alternative wordings, A, B, or D. There are three sets of models, one each for BSTART, BJOBST, and OWNMGE. The results are summarized in Appendix Table 1.

These regression models indicate that the inclusion of the time lag, in months, has no impact on the proportion responding “yes” in the BJOBST II and OWNMGE models. There is a small effect associated with the BSTART II model, with a statistically significant time lag coefficient of 0.03 per 100 per month, or 0.48 per 100 per year. This probably reflects the impact of the 1993–1996 period, as this is the only model where data are available from 1993. It is possible to estimate the impact of the time lag by comparing the BSTART I, without the time lag model, and BSTART II, with the time lag model. Adding the time lag increases the adjusted explained variance slightly, from 66% to 69%, but the F-test goodness of fit value is reduced.

Given the very small impact of the time lag on the proportion responding to one of the three screening items, it seemed appropriate to develop procedures that adjusted for screening item wording that did not incorporate any correction for the passage of time.

1.3 Development of estimates to adjust for item wording

Examination of Figs. 1 and 2 suggests that the wording of the two screening items may have had an effect on the proportion responding “yes.” As one of the major objectives of the research program is to determine the changes over time, it would be useful to have a standardized estimate of the final outcome of the item screening procedure—prevalence rates of candidate nascent entrepreneurs—adjusted for variations in item wording. The following procedure was followed to develop such estimates.
  1. (1)
    Regression analysis was used to determine the impact of the screening item wording on the proportion of “yes” responses with the use of dummy variables to represent the alternative forms, with the C and E versions the default options. The time period of the survey was NOT included in this linear model. The two linear regression models, predicting percentage of “yes” responses, were:
    $$ \begin{aligned}{} & {\text{BSTART}}\,{\text{(Yes)}} = 12.18 + ( - 4.61*A\,{\text{wording}}) + (2.26*B\,{\text{wording}}) \\ & \\ & {\text{BJOBST}}\,{\text{(Yes)}} = 6.83 + (- 2.75*D\ {\text{wording)}} \\ & \\ \end{aligned} $$
    The first model explained 66% of the variance in the proportion of “yes” responses to BSTART across 134 samples and the second 45% of the variance to BJOBST across 132 samples; both models had highly statistically significant F tests [0.0000] for goodness of fit. Item wording appears to have a major impact on the proportion that say “yes” to these items.
     
  2. (2)
    Based on the parameters in the models developed in step 1, the adjusted proportion of “yes” responses for each screening item was developed, to predict the proportion of “yes” responses that would have occurred if the wording in the PSED II screening procedure had been employed for all earlier samples; this was wording C for BSTART and wording D for BJOBST. These predictive equations were:
    $$ \begin{aligned}{} & {\text{BSTART}}\_{\text{Adjusted}} ({\text{C}}) = {\text{BSTART}}\_{\text{Actual}} + (({\text{BSTART}}\_{\text{A}})*( - 4.61)*( - 1)) + (({\text{BSTART}}\_{\text{B}})*(2.26)*( - 1)) \\ & {\text{BJOBST}}\_{\text{Adjusted}} ({\text{E}}) = {\text{BJOBST}}\_{\text{Actual}} + (({\text{BJOBST}}\_{\text{D}})*( - 2.75)*( - 1)) \\ & \\ \end{aligned} $$
    The impact would be to increase the proportion of “yes” responses to BSTART if wording A was used; reduce the proportion of “yes” responses to BSTART if wording B was used. There would be an increase in the proportion of “yes” responses to BJOBST if wording D had been employed. If the sample involved the C or E wording, the adjustments would have no effect.
     
  3. (3)
    As the PSED and GEM research programs developed there was an increase in the number of screening items from one item, in the earliest PSED pretests in 1993; to two items, for a large number of samples from 1996 through 2000; to three items, for GEM samples in 1999 and later as well as the 2004 US Assessment and the 2005–2006 PSED II screenings. As a result, two linear regression models were developed to predict the prevalence of candidate nascent entrepreneurs following the screening, one based on the two-item screening using 132 samples and the other on the three-item screening using 68 samples. The dependent variables are the proportion of respondents that answered “yes” to one or both of the two screening items (Prev_NE_2) or to any combination of three screening items (Prev_NE_3). The resultant linear models were:
    $$ \begin{aligned}{} & {\text{Prev\_NE\_2}}= -4{\text{.47}} + {{(BSTART}}*1{\text{.73)}}+{{(BJOBST}}*0{\text{.68)}} \\ & {\text{Prev\_NE\_3}} = -14.57+{\text{(BSTART}}*1.33 + {\text{(BJOBST}}*0.34) + {\text{(OWNMGE}}*1{\text{.04)}} \\ & \\ \end{aligned} $$
    The correlation between the predicted nascent 2-item prevalence rates and the actual rates was 0.92; between the predicted nascent 3-item prevalence rates and the actual rates was also 0.92. Clearly the models were good predictors of the actual outcomes.
     
  4. (4)

    The final step in the estimation procedure was to produce estimates of the candidate nascent entrepreneur prevalence rates using the estimates of the proportion of “yes” screening item responses developed in step 2 into the regression equations developed in step 3. The estimated values for BSTART and BJOBST replaced the actual values in the estimates. This would, in effect, produce estimates of the candidate nascent entrepreneur prevalence rates that would have occurred if all samples had employed screening items C, E, and F.

     
The impact is illustrated in Table 2.
Table 2

Predictions of candidate nascent entrepreneur prevalence rates with observed prevalence

 

NE predicted prevalence, 2 items, original wording

NE predicted prevalence, 3 items, original wording

NE predicted prevalence, 2 items, screening wording adjusted

NE predicted prevalence, 3 items, screening wording adjusted

Number of samples

132

68

132

68

Time span

1996–2006

2000–2006

1996–2006

2000–2006

Average value

16.5

19.9

21.3

19.9

Proportion of variance explained (R*R)

90.25%

84.64%

11.56%

73.96%

Because most of the samples with three screening items used the most recent versions, there is little impact on the estimates based on the revised wording. The three-item prevalence rates are 19.9 per 100 for both the original and revised wording estimates, and the proportion of explained variance is quite high, 85% and 74% for this set of 68 samples over 7 years. In contrast, there is considerable variation in screening item wording among the 132 samples where two items were used. The model is able to explain 90% of the variance with estimated candidate nascent entrepreneur prevalence rates but only 12% when the model uses the adjusted wording. This reflects the large number of two-item samples in PSED I project in 1998–2000. It is reasonable to conclude that adjusted item wording predictions would have a major impact on the estimated candidate nascent entrepreneur prevalence rates for these samples.

To confirm that the impact was related to item wording, and not related to changes over time, an analysis of the residuals was completed.3 If temporal change had any influence, there should be a statistically significant impact of the passage of time on the residuals from the models predicting the proportion of “yes” responses to BSTART and BJOBST. This was not the case. The correlation of time over the 1993 to 2006 period with the BSTART residuals was a non-significant 0.11 for 134 samples; time lag would explain about 1.2% of the variance in the residuals. For the 1996 through 2006 period with the BJOBST residuals the correlation was a non-significant −0.02 for 132 samples; time lag would explain about 0.04% of the variation. It is appropriate to conclude that differences in item wording were the overwhelming factors affecting variation in responses to these items, not the passage of time.

1.4 Impact of adjustments for item wording

This impact can be illustrated by presenting the two-item candidate nascent entrepreneur prevalence rates, which covers the 1998–2006 period, for each year in two formats. These are presented for the years in which data are available from 1998 to 2006 in Fig. 3. First, the actual prevalence rates developed directly from the data sets; horizontal bars represent the mean estimates of the confidence intervals represented by the vertical bars. Second, the prevalence rates estimated from the wording-adjusted predictions: circles represent mean value of estimates adjusted for the effects of the item wording on the vertical bars.
https://static-content.springer.com/image/art%3A10.1007%2Fs11187-008-9112-3/MediaObjects/11187_2008_9112_Fig3_HTML.gif
Fig. 3

Candidate nascent entrepreneur prevalence rates: original and adjusted for item wording effects: US 1998–2006

The major effect of these adjustments is to reduce the year-to-year variation and, in addition, dramatically increase the prevalence rates for the 1998–1999 period, when the substantial screening for the PSED I cohort was completed. There is, in addition, a slight reduction in the prevalence for the 2001, when the B version of the BSTART item was employed. Overall, however, there is little evidence of temporal change from 1998 to 2006.

1.5 Adjustment from candidate nascent entrepreneurs to active nascent entrepreneurs

The results are, however, the prevalence rate of those that may be considered candidates for active nascent entrepreneur status. Only those that meet three additional criteria are considered active nascent entrepreneurs. These are (1) reports of some start-up activity in the previous 12 months, (2) expectation of ownership, full or partial, of the new venture, and (3) a business initiative that has not yet reached a minimum level of financial success. As all three criteria were built into the PSED II screening interview, the rates of attrition from this project will be utilized. In these samples, 26.4% of those identified as candidate nascent entrepreneurs met the criteria to be considered active nascent entrepreneurs. That is, they reported (1) start-up activity in the past 12 months, (2) expectations of some ownership of the new firm, and (3) the absence of positive monthly cash flow that covered all expenses and salaries in 6 of the past 12 months. The results are provided in Fig. 4.
https://static-content.springer.com/image/art%3A10.1007%2Fs11187-008-9112-3/MediaObjects/11187_2008_9112_Fig4_HTML.gif
Fig. 4

Active nascent entrepreneur prevalence rates, screening item adjusted: US 1998–2006

This indicates that the prevalence rates of active nascent entrepreneurs have been relatively constant. There is no statistically significant difference between any two years during the 1998–2006 period; it has been between 5 and 6 per 100 persons 18–74 years old across this period.

These prevalence rates can be used to estimate the total number of active nascent entrepreneurs, using the total count of US citizens 18–74 years of age for each of the years (US Census Bureau 2006). The results, in Fig. 5, indicate that the total number of persons involved as active nascent entrepreneurs may have increased slightly from 10.7 million in 1998 to 11.9 million in 2006. However, a substantial proportion of this increase, about 37%, would reflect an increase in the size of the population 18–74 years of age, which was 7.4 million greater in 2006 than in 1998. There is, in summary, very little evidence of a change in the prevalence or total numbers of persons actively involved in new firm creation in the US, which is about 11–12 million persons each year. This is, however, a constantly shifting group as individuals enter and leave the firm creation experience—either by creating a new firm or disengaging from the process (Reynolds 2007).
https://static-content.springer.com/image/art%3A10.1007%2Fs11187-008-9112-3/MediaObjects/11187_2008_9112_Fig5_HTML.gif
Fig. 5

Active nascent entrepreneurs, total count and prevalence: US 1998–2006

Using a similar procedure, estimates can be developed for the prevalence of the TEA index, which combines the prevalence of active nascent entrepreneurs and those managing a firm up to 42 months in age. This is presented in Fig. 6. Note that for 1998, 1999, 2005, and 2006 only estimates—with circles at the median value–are provided. Data on survey-based prevalence rates—with cross bars representing median values–for 2004 are based on the US Entrepreneurial Assessment with a sample of 12,000, no data assembled as part of the 2004 GEM initiative in the US with a sample of 2,000 (Minniti and Bygrave 2004). Consistent with the adjusted estimates for the active nascent entrepreneur prevalence rates, there are virtually no changes over time from 1998 through 2006; the item-wording-adjusted US TEA index prevalence rate is constant at about 10 per 100 persons 18–74 years of age.
https://static-content.springer.com/image/art%3A10.1007%2Fs11187-008-9112-3/MediaObjects/11187_2008_9112_Fig6_HTML.gif
Fig. 6

Prevalence rates of TEA active entrepreneurs: original and adjusted for item wording effects: US 1998–2006

There are three other programs in the US that have developed measures designed to provide longitudinal or time-series measures of the new firm creation process in the US. The US Bureau of the Census (Census), in collaboration with the US Small Business Administration track the initial registrations of new employee firms (single or multi-establishment), based on initial filing of federal social security payments; data are available from 1990 to 2004 (US Small Business Administration 2004). The US Bureau of Labor Statistics (BLS) tracks establishments that provide initial state unemployment insurance payments; data are available from 1992 to 2003 (Spletzer et al. 2004). Analysis of the US Current Population Surveys (CPS), where 130,000 individuals are interviewed each month, are utilized to identify those reporting a substantial month-to-month increase in attention to business ownership, now called the Kauffman Entrepreneurial Index; data are available from 1996 through 2005 (Fairlie 2006). It should be noted that the unit of analysis is different for all three measures. The CPS measure reflects reports of individual human respondents. The BLS measure reflects new registration of a business establishment, a single location where business activity takes place. The Census measure is related to new registrations of a business entity, either single or multi-establishment. For this analysis they are all converted into a prevalence rate based on the adult population, although there is some variation in the age range utilized in the calculations.

The patters of all four measures, using the item-wording-adjusted measures from the nascent entrepreneur research programs, are presented in Fig. 7. All indicators of firm creation activity indicate no substantial change in the prevalence rates over the time periods for which data are available. The higher year-to-year variation in the nascent entrepreneur measures probably reflects the smaller sample sizes for this program.
https://static-content.springer.com/image/art%3A10.1007%2Fs11187-008-9112-3/MediaObjects/11187_2008_9112_Fig7_HTML.gif
Fig. 7

Four measures of US business creation: 1990–2005

By adjusting the vertical axis to a logarithmic scale, the year-to-year changes have the same vertical adjustments—a 10% change with a base of 10 would have the same vertical shift as a 10% change on a base of 1 or 0.1. Figure 8 presents the same information as that in Fig. 7 with such a scale, and makes it even more obvious that the prevalence rates are remarkably consistent over the 1990 to 2005 period—this is particularly true for the TEA rate, the top line.
https://static-content.springer.com/image/art%3A10.1007%2Fs11187-008-9112-3/MediaObjects/11187_2008_9112_Fig8_HTML.gif
Fig. 8

Four measures of US business creation, log10: 1990–2005

The consistency across the four measures increases confidence in the importance of making screening item adjustments to the current programs identifying nascent entrepreneurs. It also suggests that these four US efforts are indicators of different features of the same firm creation processes. The differences among the measures reflect the fact that they are identifying new business creation at different stages of the process.

2 Conclusions and implications

The major finding is straightforward. Differences in wording of screening items can have a major effect on the prevalence of candidate nascent entrepreneurs. Adjustments for the number and wording of screening items doubled the estimated prevalence rate of candidate nascent entrepreneurs in the 1998–1999 period, from 11 to 22 per 100. Adjustments for screening item effects lead to no temporal change in the prevalence of active nascent entrepreneurs, at 5–6 per 100, over the 1998–2006 period. In the absence of adjustments to compensate for item wording effects, misleading inferences about time-series patterns may occur. This appears to reflect the sensitivity of procedures locating relatively rare activity in a sample representative of the entire adult population.

This can affect two important research objectives. First, temporal comparisons in the same country with the same language require careful attention to the potential effect of variation in screening item wording. The analysis of US data makes clear that major errors of temporal patterns may occur if screening item wording is not taken into account.

Implications for cross-national comparisons, the second major research objective, are equally significant. In the GEM initiative, for example, the initial schedule is developed in English and then translated into the various languages used in different national surveys. Given the sensitivity to variations in screen items in English, there may be major differences when the items are translated into another language.

One solution to this complication would be for all human population surveys to locate nascent entrepreneurs to use a common set of screening items. Three such items—identified as C, E, and F in this analysis—are now being employed in a wide range of human population surveys. They have been utilized in the second US PSED project, are to be utilized in a major panel study currently being implemented in Australia (the Comprehensive Australian Study of Entrepreneurial Emergence),4 and have become widely utilized in the various national surveys that are part of the GEM research program. This greatly facilitates the potential for comparisons among these different initiatives. There is now substantial experience with utilizing these three items and the results are well documented. While there is no “one best way” to define or locate nascent entrepreneurs, this “three item standard” seems a good solution.

It is possible that the yield of candidate nascent entrepreneur may increase if additional screening items were included. More items may have little effect on high prevalence countries, where prevalence estimates are in excess of 20 per 100 in the population; an increase in 3–5 per 100 respondents would not have a major impact on these assessments. But in countries with very low prevalence rates, below 5 per 100, and increase of 1–2 per 100 respondents can have a major effect on interpretations. Further, there has not been much attention on the impact of translation of these three items into other languages. To fully convey the idea of a new business that encompasses self-employment and selling—or trading—goods or services may require additional items in some languages. As long as the three basic items are included, however, there will be some potential for comparisons with other samples based on interviews in other languages. More systematic research on the impact of screening items on estimates of nascent entrepreneur prevalence rates could be very useful.

Footnotes
1

Most GEM global and national reports are available at ‘www.gemconsortium.org.’ The author had primary responsibility for five global reports, listed in the references, during his service as founding coordinating principal investigator from 1998–2003 (Reynolds et al. 1999, 2000, 2001, 2002).

 
2

Most data are, or will be, in the public domain. Data from PSED Projects available at ‘www.psed.isr.umich.edu.’ The 2004 US Assessment is available from the ICPSR archives at the U Michigan as project 4688, the consolidated. GEM adult population national data sets for 1999–2003 are available at ICPSR as project 20320.

 
3

After a review of a preliminary version, Per Davidsson suggested this assessment.

 
4

The Australian CAUSEE project, with Per Davidson of Queensland University of Technology as Coordinating Principal Investigator, has, as of February 2008, completed most of the initial screening and first wave of detailed data collection.

 

Acknowledgments

I am indebted to Per Davidsson, Zoltan Acds, Richard Curtain, and Scott Shane for their helpful comments on preliminary drafts.

Copyright information

© Springer Science+Business Media, LLC. 2008