HOW DO WE REACH THEM? COMPARING RANDOM SAMPLES FROM MOBILE AND LANDLINE PHONES. Paul Steffens* Per Davidsson* Marcello Tonelli* Michael Stuetzer*

Transcription

1 HOW DO WE REACH THEM? COMPARING RANDOM SAMPLES FROM MOBILE AND LANDLINE PHONES Paul Steffens* Per Davidsson* Marcello Tonelli* Michael Stuetzer* * Australian Centre for Entrepreneurship Research, QUT Business School, Queensland University of Technology, Australia. 1

2 HOW DO WE REACH THEM? COMPARING RANDOM SAMPLES FROM MOBILE AND LANDLINE PHONES Abstract Quantitative studies of nascent entrepreneurs such as GEM and PSED rely on screening the adult population to generate a representative sample, usually by phone in developed economies. However, phone survey research has recently been challenged by shifting patterns of ownership and response rates of landline versus mobile (cell) phones, particularly for younger respondents. This challenge is acutely intense for entrepreneurship which is a strongly age-dependent phenomenon. Although shifting ownership rates have received some attention, shifting response rates have remained largely unexplored. For the Australian GEM 2010 adult population study we conducted a dual-frame approach that allows comparison between samples of mobile and landline phones. We find a substantial response bias towards younger, male and metropolitan respondents for mobile phones far greater than explained by ownership rates. We also found these response rate differences significantly biases the estimates of the prevalence of early stage entrepreneurship by both samples, even when each sample is weighted to match the Australian population. Our results underscore the importance of using appropriate survey modes to ensure representativeness of samples used in entrepreneurship research. 1 Introduction Many studies in entrepreneurship, including prominent studies of nascent entrepreneurship such as the Global Entrepreneurship Monitor (GEM) and the Panel Study of Entrepreneurial Dynamics (PSED), rely on obtaining representative samples of households or individuals. Identifying a representative sample of the population is particularly crucial for GEM and other studies that have a primary interest in the prevalence rate of entrepreneurial phenomena. This is normally done using a phone screening interview in developed economies. However, phone survey research that relies on reaching a random sample of the adult population has recently been challenged by shifting patterns of ownership and response rates of landline (or fixed-line) phones versus mobile phones (also called cell or wireless phones) especially for younger respondents (Blumberg and Luke 2007a; 2010; Ehlen and Ehlen 2007; Keeter et al. 2007). This is particularly salient for studies in entrepreneurship which is a highly agedependent phenomenon (Bosma and Levie 2010). The issue of an increasing mobile-only population and its impact on the representativeness of samples generated from landline only studies, has received some attention in the broader literature (Tucker at al. 2007; Link et al. 2007; Lavrakas et al. 2007; Mokrzycki et al. 2009; Voigt et al. 2011). In response, for example, GEM requires 2

3 a secondary mobile-only sample if the landline coverage in a country is lower than 80%. However, what has received almost no attention is the variation in nonresponse bias between mobile and landline phones. Yet two studies by Brick et al. (2006) and Kennedy (2007) reveal strong response rate variations for mobile and landline phones, with systematic variations by age. If the same pattern will be observed in other samples, a severe problem arises for entrepreneurship studies such as GEM and PSED. Nonresponse bias impairs the representativeness of the underlying samples and undermines the robustness of conclusions based on comparisons of entrepreneurship prevalence rates across time and nations. To investigate the impact of nonresponse bias for mobile versus landline phones on estimates of prevalence of entrepreneurial phenomena, we conducted a dual-frame sample of mobile and landline phones for the 2010 Australian GEM adult population survey. Two random samples of 1,000 adults were generated by random-digitdialling fixed landline and mobile phone numbers, respectively, across Australia. By comparing the sub-sample of individuals who own both types of phones generated from each sample, we can determine the level of nonresponse bias. We compare their demographic composition in terms of age, gender and metro/non-metro residence. We also compare the estimates of the prevalence rates of early stage entrepreneurship obtained from each sample. In doing so, this paper makes an important method contribution by assessing sampling techniques often used in entrepreneurship research. 2 Nascent entrepreneurship research New firm creation represents a central focus of empirical research in entrepreneurship (Davidsson 2004) and estimating the prevalence of nascent entrepreneurs has recently become a critical objective of several prominent studies. The first systematic research on the pre-operational stage of business creation was conducted in Wisconsin in 1992 (Reynolds and White 1997), when the term nascent entrepreneur was first coined (Reynolds and Miller 1992), referring to an individual engaged in an on-going but not yet operational business start-up. This endeavor was the precursor of the first US PSED I (Gartner et al. 2004) and subsequently the GEM initiative, which applied the PSED sampling technique (Davidsson 2005). 3

4 GEM is a large, policy-orientated research initiative that aims to compare and assess the prevalence rate of entrepreneurial activity between countries. GEM aims to investigate the prevalence rate of the adult population involved in business start-ups (nascent entrepreneurs), early stage business, and other entrepreneurial activities. Hence it is particularly salient for GEM to generate a sample that is as close as possible to a representative, random sample of the adult population. Both GEM and PSED use random sampling of households because this early stage of entrepreneurship is not captured on any lists of businesses or companies. The most cost effective way of screening a representative sample of the adult population in developed economies is through phone surveys. For GEM 2010, phone surveys were employed in 35 countries. Of those, 12 countries collected data from a primary sample of landlines together with a secondary sample of mobile phones. 3 Noncoverage and nonresponse bias from landline and mobile samples A shift from landline to mobile telephone ownership and usage has been increasing over time (Kuusela et al. 2008), challenging researchers to question the representativeness of samples from landline only samples, and the bias of population estimates. In particular, the mobile-phone only segment of the population has been examined by several studies. Estimates from the National Health Interview Survey show that the number of mobile-only households has increased rapidly from 4.4% in 2004 to 13.6% in 2007 to 26.6% in This trend is especially salient for younger respondents, where the mobile-only percentage has risen from 10.1% in 2004 to 45.2% in 2010 (Blumberg and Luke 2007a; 2007b; 2010; Tucker et al. 2007). The shift in the use of communication techniques can lead to the well-known noncoverage bias (Dillman 2007). This bias occurs in landline only samples if covered (landline only and dual user) households and individuals differ from noncovered (non-landline) households and individuals in characteristics under investigation. A number of studies have investigated the impact of noncoverage bias of mobile-only respondents when using landline 4

5 samples. Overall, results indicate that for many variables of interest, overall population estimates are not substantially biased, but estimates for younger subgroups of the population are. For example, in a study of media use, political and social attitudes conducted in 2006, Keeter et al. (2007) found that population estimates of most variables were not significantly different between a landline and a blended landline / mobile sample. However, they were often significantly different for population estimates of younger population sub-groups. In contrast, comparing estimates from a landline and face-to-face interviews in 2006, Blumberg and Luke (2007) showed significant differences for several health indicators and attitudes. Similarly, Link et al. (2007) found significant differences between landline only and combined landline and mobile samples for 3 of 10 health indicators. An exit poll in the 2008 November election revealed voter preference differences between cell-only and landline accessible voters, including older (over 30) voters (Mokrzycki at al. 2009). Taken as a whole, three things are clear from these studies: 1 Estimates of younger sub-groups of the population which have higher rates of mobile-only tend to be biased from landline only studies; 2 Estimates of variables for the entire population that demonstrate age-variation tend to be biased from landline only studies; 3 These biases will continue to increase over time as mobile-only rates increase. Importantly, since entrepreneurial behaviour is strongly age dependent (Bosma and Levie 2010), it is likely that population estimates will be affected by the noncoverage bias associated with mobile-only individuals. While several studies have investigated this noncoverage issue, we identified only two studies dealing with the issue of nonresponse biases of landline versus mobile phones. Brick et al. (2006) conducted a dual frame survey of mobile and landline numbers in 2004, and compared some results with a face-to-face survey over the same time period to evaluate nonresponse bias. They demonstrate a strong response variation: according to the face-to-face survey, 53.4% of those owning a landline reported owning a mobile phone, compared with 71.0% of the landline sample. Hence, those with mobile phone are strongly over-represented in the landline sample. Conversely 88.5% of 5

6 households with a mobile phone in the face-to-face survey reported owning a landline, compared with 77.5% of the mobile sample. Using a different dataset, Kennedy (2007) replicated most of these findings. This response variation can result in a nonresponse bias if respondents and nonrespondents in the survey samples differ in important characteristics (Dillmann 2007; Groves 2006) that influence their decision to engage in entrepreneurship. Brick et al. (2006) demonstrate that response variations in their samples were highly age dependent. As a consequence, a separate sample to access the mobile-only individuals is not sufficient to eliminate response biases (Brick et al. 2007). Furthermore, other studies found differences between landline and mobile surveys with regard to gender (Link et al. 2007; Kennedy 2007) and residence (metropolitan vs. non-metropolitan regions; Tucker et al. 2007) which are important determinants of entrepreneurial activity. If variations in response rate by age, gender, and residence holds true in our sample, population estimates of entrepreneurial activity may still be biased when a mobile sample is included. 4 Data collection 4.1 The GEM study The paper reports analysis of the GEM adult population survey for Australia in The Global Entrepreneurship Monitor (GEM) research program is an annual assessment of the national level of entrepreneurial activity. It was initiated in 1999 with 10 countries, expanded to 21 in the year 2000, with 29 countries in 2001 and 37 countries in GEM 2009 conducted research in 56 countries. Every year each national team is responsible for conducting a survey of at least 2,000 people within its adult population. The Adult Population Survey (APS) is a survey of attitudes towards entrepreneurship in the general population but it also asks people whether or not they are engaged in start-up activity or own or run a business. 6

7 4.2 Sampling approach The Australian 2010 APS was a telephone survey of 2,000 adults 18 years or older. There is a very high coverage for both mobile and landline telephones in Australia (>80%). We generated two equal sized random samples from landline (n=1,000) and mobile phones (n=1,000) respectively. For fixed lines, the respondent was the adult with the next birthday in the household. For mobile phones, the respondent was the main user of this phone. The sample framework was the universe of all telephone numbers. The sample was selected using random digit dialling (full RDD for the fixed lines and generate the mobile phone numbers using Australian Communications and Media Authority (ACMA) mobile phone prefixes as the stem, and randomly-generating the remaining numbers). A phone schedule was developed to include a mix of weekday, week night and weekend calls. There was a minimum of five call-backs to each number if it wasn t answered. For landline numbers, the interviewer asked to speak to the adult in the household with the next birthday. If they were not home, again a minimum of five call-back attempts were made to talk with this individual. 4.3 The GEM measures The APS uses a harmonized questionnaire across all participating countries. The survey measures a prevalence rate of entrepreneurial activity, and for those who are involved asks them some questions about the nature of their business and the business environment (Reynolds et al. 2005). The measure of entrepreneurial activity is the Early Stage Entrepreneurial Activity prevalence rate (also called TEA index). This indicator is calculated in an identical way in each country. Respondents are asked three questions that form the basis of the TEA index: 7

8 1. Are you, alone or with others, currently trying to start a new business, including any selfemployment or selling any goods or services to others? 2. Are you, alone or with others, currently trying to start a new business or a new venture for your employer as part of your normal work? 3. Are you, alone or with others, currently the owner of a business you help manage, self-employed or selling any goods or services to others? Those who respond positively to these questions are also asked filter questions to ensure they are actively engaged in business creation as owners and managers, how long they have been paying wages to employees, and other questions about motives and time to start up, sources of finance and numbers of jobs created. A distinction is made between two types of entrepreneurs: nascent entrepreneurs (those that have been paying salaries for less than three months) and new business owner-managers (those that have been paying salaries for between three and 42 months). Early-Stage Entrepreneurial Activity is the sum of the nascent entrepreneurs and baby business owner/managers minus any double counting (i.e. those who respond positively to both). The Early-Stage Entrepreneurial Activity rate is computed by dividing the number of entrepreneurs by the number of respondents. This rate is comparable across nations and measures the propensity of a country to be entrepreneurial. Since we used mixed survey modes, we employed additional two questions. Individuals selected from the landline sample were asked whether or not they also owned a mobile phone, while individuals selected from the mobile sample were asked if they also had a landline number (including VoIP phone used only for incoming calls). This information gave us the ability to adjust through weighting if necessary. 5 Analysis and results Below we provide background to phone ownership trends within our empirical setting Australia. Informed by this, we conduct three types of analyses to compare the mobile and landline phone samples. First, to examine the existence of the nonresponse bias, we compare important demographic variables between those respondents who 8

9 have both types of phones but either are part of the mobile or the landline sample. Second, we examine the impact of the nonresponse and noncoverage bias on our samples. To this end we compare the demographics characteristics of the mobile and landline samples, and also compare these with the actual Australian population figures. Third, to evaluate the impact of biases on the estimates of entrepreneurial activity prevalence rates, we estimate the TEA index from each sample and compare these with the combined sample. We do this for both the unweighted (original data) and for each sample weighted to match the Australian population (for age, gender and metro/non-metro). For all analyses, we test for significant differences between the samples. 5.1 Changing telephone ownership patterns in Australia Similar to the US, the Australian communications environment is rapidly changing, providing consumers with a range of alternatives. Figure 1 displays the trend in the percentage of individuals with ownership or access to mobile and landline services across Australia. Thereby we draw on data obtained from the Roy Morgan Single Source Survey (RMSSS), a large survey of 20,000 households that are enrolled face-to-face, as reported in a government report (ACMA 2009). While landline phones remain the most common choice in terms of voice minutes, Australians are increasingly turning to mobile technology to make voice calls and use their landline service solely to maintain an internet connection. This trend anticipates a surpass of mobile traffic in the near future (ACMA 2009). ******************************* INSERT FIGURE 1 ABOUT HERE ******************************* Since June 2007, the number of mobile services operating in Australia has risen by four per cent to million. This increase appears to be linked with changes in digital communication services. The closure of the code-division multiple access (CDMA) network in 2008 led to nearly 40 per cent of mobile users to subscribe to a 3G service, indicating the potential for landline users to cancel their service if intended solely as a mean to have an internet connection. 9

10 While the process of moving from landline phones to mobiles is in progress across all age groups of Australians, the ACMA (2009) survey suggests a strong correlation between the age of the consumer and substitution: younger Australians are leading the shift away from landline communications, while elders remain generally more attached to landline technology. Figure 2 indicates a strong relationship between age and ownership of mobiles and fixed-line services. As illustrated, older Australians are more likely to have a fixed-line phone, with 94 per cent of those aged 70 and over maintaining a home fixed-line phone service as opposed to just 52 per cent having a mobile. Conversely, young Australians show a stronger preference towards mobile communications technology. In fact just 75 per cent of 18 to 24-years-old have a fixed-line service, while 92 per cent have mobiles. The proportion of mobile-only is high for younger individuals (33% for and 18% for 25-34), but much lower for older age groups. ******************************* INSERT FIGURE 2 ABOUT HERE ******************************* 5.2 Response rate variations: Comparison of demographic variables in the two samples The first step in our empirical analysis is to compare those 1,464 respondents who have both types of phones but either belong to the mobile sample (616 respondents) or the landline sample (848 respondents). If there exists a nonresponse bias, significant differences between both groups should be identifiable. Figures 3 to 5 give a first impression of these differences for age, gender, and metro/non-metro residence of the respondents. Regarding age (Figure 3) the data clearly show that younger individuals are more likely to respond to mobile phone calls, while older respondents are more likely to respond to landline calls. A Chi-squared test (Chi-square 120.7; d. f. 5; p<0.001) confirms that these differences are significant. Turning to gender, Figure 4 indicates that men are more likely to respond to mobile calls while women rather respond to landline calls. Again, these differences are statistically significant (Chi-square 22.1; d. f. 1; p<0.001). Figure 5 shows that in metropolitan regions individuals are more likely to respond to mobile calls, while in non-metropolitan regions individuals are more likely to be reached by landline. Employing aa Chi-square test (Chi-square 16.8; d. f. 1; p<0.001) underpins this observation. 10

11 Taken together, this step of the analyses clearly confirms a substantial nonresponse bias, with nonresponse that differs differing between mobile and landline samples. ******************************* INSERT FIGURE 3-5 ABOUT HERE ******************************* 5.3 Comparison of demographics of the samples In a second step, we analyse the implications of that nonresponse bias and the noncoverage bias on the demographic profiles of our both samples: mobile and landline. We compare the age distributions, gender and metro/non-metro residential location of each sample with the combined sample and with the population statistics for Australia. Figure 6 gives a first impression of the results regarding the age distribution. It is immediately evident that while the combined sample closely matches the Australian population, the mobile sample clearly has a substantially higher percentage of the younger respondents, and conversely the landline sample is heavily biased towards older respondents. Table 1 displays the percentages of each sample in each age band. We employ z-tests for the proportion of each sample falling into an age band against the population proportion for Australia. These tests confirm that most age bands are significantly different to the Australian population for both the mobile and landline samples. However the combined sample matches the true population reasonably well. Only the youngest age category is slightly over-represented (1.6% p<0.05) and the oldest age category is slightly under-represented (3.3%; p<0.001). In Figure 7 we present the gender balance of each sample, the combined sample and the actual population figure. This figure suggests that males are overrepresented in the mobile sample while females are overrepresented in the landline sample. The combined sample is again a closer match to the Australian population although females 11

12 are slightly overrepresented. Table 2 reveals that the mobile, landline and the combined samples are all significantly different to the Australian population. Figure 8 displays the metro/non-metro balance of each sample, the combined sample and the actual population figure. It is clear that metropolitan regions are overrepresented in the mobile sample and nonmetropolitan regions are overrepresented in the landline sample. The combined sample is again a very close match to the Australian population. Table 3 shows that while both the mobile and landline samples are significantly different to the Australian population, the combined sample is not. ************************************************** INSERT FIGURES 6-8 & TABLES 1-3 ABOUT HERE ************************************************** 5.4 Estimates of entrepreneurial prevalence rates Finally we turn our attention to assessing the impact of the sample biases on the estimation of entrepreneurial activity prevalence rates. First we estimate the TEA index, and its two sub-components nascent entrepreneurship rate and early stage business owner-manager rate. Our first analysis shows a naïve estimate without any weighting applied to match the demographics of the population. The second analysis weights each of the samples to the Australian population as the GEM study does. We used three demographic variables to generate the weights age, gender and metro / non-metro residence. We did not have age data for 18 of the respondents, 9 from each sample. These respondents were not used for the weighted analysis. Table 4 and Table 5 shows the point estimates and 0.05 confidence interval for each sample for the unweighted and weighted analysis respectively. Z tests indicate whether the landline and mobile samples are significantly different. An initial impression of the results for the weighted estimates can be gained from Figure 9 where the rates are displayed. The mobile sample provides higher estimates of the TEA index and its subcomponents than the landline sample. Table 4 confirms that when the data is not weighted that this difference is significant for the TEA index (8.3% vs 5.4%; p<0.01), and significant at the 0.1 level for the prevalence rate of early 12

13 stage business owner managers. The difference is not significant for the prevalence rate of nascent entrepreneurs, but this is most likely due to the low power of the test the sample sizes of 1,000 each is not quite large enough to reveal difference in the low prevalence rate in the range of 2-4%. Importantly, it can be seen from Table 5 that weighting each sample to the Australian population in attempt to correct for the observed biases in age, gender and metro/non-metro does little to improve the bias in the estimates of entrepreneurial activity prevalence. The TEA remains significantly higher for the mobile sample than the landline sample (7.6% vs 5.2%; p<0.01), with the difference between the two estimates only marginally reduced from 2.9% to 2.4%. ************************************************** INSERT FIGURE 9 & TABLES 4-5 ABOUT HERE ************************************************** 6 Discussion and conclusions Although the issue of noncoverage bias of both landline and mobile phone samples is well established (Tucker et al., 2007), non response bias has been investigated by only one earlier study (Brick et al. 2006). By comparing the demographic profiles of respondents who own both a landline and mobile phone generated from a sample contacted by landline phone versus a sample generated by mobile phone, we demonstrate that there are systematic differences in response rates between mobile and landline phones according to the age, gender and metro/non-metro residence of respondents. Younger, male and individuals who reside in metropolitan location are more likely to respond on mobile phones, with the reverse being true for landline phones. Regarding age, thisthis complements the earlier findings from of Brick et al. (2006) who report that older participants were more likely to respond to landline calls. We could also replicateour results are also consistent with those of the result from Kennedy (2007) that men are more likely to respond to mobile calls. We find that a dual-frame sample of 50% landlines and 50% mobile phones successfully generates a sample that is as close as possible to a representative random sample of the population. Our comparison of the 13

14 demographic composition of the landline and mobile phone samples reveal substantive and significant differences compared with the Australian population. In particular, the age distribution of the two samples differed markedly far more than expected by the previously reported ownership rates. For example, the age group represented 24% of the mobile sample, but only 5% of the landline sample. In contrast, the over 65 age group represented 25% of the landline sample, but only 5% of the mobile sample. Our analysis also revealed that either sample alone would have produced biased estimates of the prevalence rates for involvement in start-up ventures. This is true even if each sample was weighted to match the demographic profile of Australian adults according to age, gender and metro/non-metro following the methodology for the GEM study. Prevalence rates were overestimated from mobile phones and underestimated from landline phones. It appears that in Australia, using a combination of mobile and landline phones is likely to yield a more representative sample of individuals. The combined sample matched the age and the geographic distribution of the Australian population very closely. Hence this study further validates the notion that it is becoming increasingly challenging to generate a reasonably representative sample through telephone surveys as the ownership, and even more importantly usage patterns of landline and mobile phones is rapidly evolving. However, understanding these patterns can help to design a mixed-modes survey approach to obtain the representativeness that would otherwise be missed (see for a discussion of survey modes and the associated advantages and disadvantages of mixed modes de Leeuw and Van Der Zowen 1988; Dillmann 2002; Dillmann and Christian 2005; Fowler et al. 1998; Tortora et al. 2008). More specifically, in the case of which individuals are more likely to answer their landline or mobile phones, this research determined that younger people are more prone to answer their mobiles, while elders illustrate a similar trend for landlines. This research has implications for all telephone-based survey research that aims to generate a random sample of individuals. However it is particularly relevant for research that intends to identify the prevalence rate of entrepreneurial activity in the population such as the GEM research project. First of all, not correcting for noncoverage and nonresponse bias will impair the representativeness of the generated samples. In Australia an equal 14

15 split of the sample in landline and mobile phones fixed the noncoverage bias. As coverage and use of phones are dynamically evolving this ratio is likely to change in the future. Future research should particularly focus on the nonresponse bias which is harder to both estimate and overcome (see for a discussion of several approaches Brick et al. 2006). Second, given the long-term worldwide trend of increasing mobile coverage and use (Kuusela et al. 2004), it is likely that the prevalence rate of entrepreneurial activity is underestimated by those studies which rely on landline-only samples. Third, researchers should be cautious when comparing prevalence rates across time and nations, which are two important research objectives of GEM (Reynolds 2009). Such differences in prevalence rates might easily be attributable to these biases which probably also differ across time and geographic location. We are aware that it is easier said than done to account for these biases. However such efforts would increase the validity of studies employing such a research design. 15

16 References ACMA (2009). Convergence and communications Report 1: Australian household consumers take-up and use of voice communications services. Australian Communications and Media Authority, Commonwealth Government, Australia. Available online at Bosma, N., & Levie, J. (2010). Global Entrepreneurship Monitor: 2009 Global Report. Global Entrepreneurship Research Association, Boston MA. Blumberg, S. J., & Luke, J. V. (2007a). Coverage bias in traditional telephone surveys of low-income and young adults. Public Opinion Quarterly, 71(5), Blumberg, S. J., & Luke, J. V. (2007b). Wireless substitution: Early release of estimates from the national health interview survey January to June, National Center for Health Statistics, USA. Available online at Blumberg, S. J., & Luke, J. V. (2010). Wireless substitution: Early release of estimates from the national health interview survey January to June, National Center for Health Statistics, USA. Available online at Brick, M. J., Dipko, S., Presser, S., Tucker, C., & Yuan, Y. (2006). Nonresponse bias in a dual frame sample of cell and landline numbers. Public Opinion Quarterly, 70(5), Brick, M. J., Dipko, S., Presser, S., Tucker, C., & Yuan, Y. (2007). Cell phone survey feasibility in the U.S.: Sampling and calling cell numbers versus landline numbers. Public Opinion Quarterly, 71(1) de Leeuw, E., & Van Van Der Zowen, H. (1988). Data quality in telephone and face-to-face surveys: A comparative analysis. In R. M. Groves, P. Biemer, L. Lyberg, J. T. Massey, W. L. Nicholls II & J. Waksberg (Eds.), Telephone survey methodology (pp ). New York: Wiley-Interscience. Davidsson, P. (2004). Researching entrepreneurship. NY: Springer. Davidsson, P. (2005). Paul D. Reynolds: Entrepreneurship research innovator, coordinator, and disseminator. Small Business Economics, 24(4), Dillman, D. A. (2002). Navigating the rapids of change: Some observations on survey methodology in early twentyfirst century. Public Opinion Quarterly, 66(3),

17 Dillman, D. A. (2007). Mail and internet surveys: The tailored design method (2nd ed., updated.). Hoboken, NJ: John Wiley & Sons. Dillman, D. A., & Christian, L.M. (2005). Survey mode as a source of instability in responses across surveys. Field Methods, 17(1), Ehlen, J., & Ehlen, P. (2007). Cellular-only substitution in the United States as lifestyle adoption: Implications for telephone survey coverage. Public Opinion Quarterly, 71(5), Fowler, F. J., Jr., Roman. A. M., & Di, Z. X. (1998). Mode effects in a survey of Medicare prostate survey patients. Public Opinion Quarterly, 62(1), 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. Groves, R. M. (2006). Nonresponse rates and nonresponse bias in household surveys. Public Opinion Quarterly, 70(5), Keeter, S., Kennedy, C., Clark, A., Tompson, T., & Mokrzycki, M. (2007). What s missing from national landline RDD surveys? The impact of the growing cell-only population. Public Opinion Quarterly, 71(5), Kennedy, C. (2007). Evaluating the effects of screening for telephone service in dual frame RDD surveys. Public Opinion Quarterly, 71(5), Kuusela, V., Callegaro, M., & Vehovar, V. (2008). The influence of mobile telephones on telephone surveys. In J. M. Lepkowski, C. Tucker, J. M. Brick, E. De Leeuw, L. Japec, P. J. Lavrakas, M. W. Link & R. L. Sangster (Eds.), Advances in telephone survey methodology (pp ). Hoboken, New Jersey: Wiley- Interscience. Lavrakas, P. J., Shuttles, C. D., Steeh, C., & Fienberg, H. (2007). The state of surveying cell phone numbers in the United States: 2007 and beyond. Public Opinion Quarterly, 71(5), Link, M. W., Battaglia, M. P., Frankel, M. R. Osborn, L., & Mokdad, A. H. (2007). Reaching the U.S. Cell Phone Generation. Public Opinion Quarterly, 71(5), Mokrzycki, M., Keeter, S., & Kennedy, C. (2009). Cell-phone-only voters in the 2008 exit poll and implications for future noncoverage bias. Public Opinion Quarterly, 73(5),

18 Reynolds, P. D. (2009). Screening item effects in estimating the prevalence of nascent entrepreneurs. Small Business Economics, 33(2), Reynolds, P. D., Bosma, N., Autio, E., Hunt, S. De Bono, N., Servais, I. Lopez-Garcia, P., & Chin, N. (2005). Global Entrepreneurship Monitor: Data Collection Design and Implementation Small Business Economics, 24(3), Reynolds, P., & Miller, B. (1992). New firm gestation: Conception, birth, and implications for research. Journal of Business Venturing, 7(5), Reynolds, P. D., & White, S. B. (1997). The entrepreneurial process: Economic growth, men, women, and minorities. Westport, CT: Quorum Books. Tortora, R., Groves, R. M., & Peytcheva, E. (2008). Multiplicity-based sampling for the mobile telephone population: Coverage, nonresponse, and measurement issues. In J. M. Lepkowski, C. Tucker, J. M. Brick, E. De Leeuw, L. Japec, P. J. Lavrakas, M. W. Link & R. L. Sangster (Eds.), Advances in telephone survey methodology (pp ). Hoboken, New Jersey: Wiley-Interscience. Tucker, C., Brick, J. M., Meekins, B. (2007). Household telephone service and usage patterns in the United States in 2004: Implications for telephone samples. Public Opinion Quarterly, 71(1), Voigt, L. F., Schwartz, S. M., Doody, D. R., Lee, S. C. and Li, C. I. (2011), Feasibility of including cellular telephone numbers in random digit dialing for epidemiologic case-control studies, American Journal of Epidemiology, 173(1),

19 Figure 1: Ownership / Usage of Landline and Mobile Phones Source: ACMA (2009) 19

20 Ownership Figure 2: Ownership of Mobile and Landline Phones by Age Band 120% 100% 92% 95% 89% 94% 96% 94% 80% 60% 40% 20% 0% 90% 80% 81% 75% 71% Mobile Phones 33% 18% 7% Landline Phones 52% Mobile Only 3% 2% 1% and over Age Source: ACMA (2009) 20

21 Percentage of Sample Figure 3: Comparing Respondents Who Own Both Mobiles and Landlines: Age 25% 20% 15% 10% 5% Mobile Sample Landline Sample 0% >65 Age Group 21

22 Percentage of Sample Figure 4: Comparing Respondents Who Own Both Mobiles and Landlines: Gender 70% 60% 50% 40% 30% 20% 10% 0% Male Gender Female Mobile Sample Landline Sample 22

23 Percentage of Sample Figure 5: Comparing Respondents Who Own Both Mobiles and Landlines: Residence 80% 70% 60% 50% 40% 30% 20% Mobile Sample Landline Sample 10% 0% Metro Non-metro 23

24 Percentage of Total Figure 6: Comparing Age Distribution of Samples with the Australian Population 30% 25% 20% 15% 10% 5% 0% Age Bracket Population Landline Sample Mobile Sample Combined Sample 24

25 Percentage of Total Figure 7: Comparing Gender Balance of Samples with the Australian Population 70% 60% 50% 40% 30% 20% 10% 0% Population Mobile Sample Landline Sample Combined Sample Male Female 25

26 Percentage of Total Figure 8: Comparing Metro/Non-Metro Balance of Samples with the Australian Population 80.0% 70.0% 60.0% 50.0% 40.0% 30.0% 20.0% 10.0% 0.0% Population Mobile Sample Landline Sample Combined Sample Metro Non-Metro 26

27 Percentage of Population Figure 9: Weighted Estimates of Entrepreneurial Activity Prevalence Rate for Each Sample 8.0% 7.0% 6.0% 5.0% 4.0% 3.0% 2.0% 1.0% 0.0% Nascent Entrepreneurs Early Stage Business Owner Managers Total Early Stage Entrepeurial Activity Mobile Sample Landline Sample Combined Sample 27

28 Table 1: Comparing Age Bands of Samples with the Australian Population Mobile Landline Combined Australian Sample Sample Sample Population Column N % Column N % Column N % Column N % Age Bands *** 5.8*** 14.7* *** 10.9*** * * 21.9*** *** 19.7*** *** 23.3*** 14.0*** 17.3 Test of difference between mobile and landline samples: Chi-square (d.f. 5) *** Table displays z-test of proportion of sample compared with Australian population *** p<.001; ** p<0.01; * p<

29 Table 2: Comparing Gender Balance of Samples with the Australian Population Mobile Landline Combined Australian Sample Sample Sample Population Column N % Column N % Column N % Column N % Male 54.4*** 38.5*** 46.5** 49.3 Female 45.6*** 61.5*** 53.6** 50.7 Test of difference between mobile and landline samples: Chi-square (d.f. 1) 50.8*** Table displays z-test of proportion of sample compared with Australian population *** p<.001; ** p<0.01; * p<

30 Table 3: Comparing metro/non-metro Balance of Samples with the Australian Population Mobile Combined Australian Sample Landline Sample Sample Population Column N % Column N % Column N % Column N % Outside Capital City 30.1*** 40.8*** Capital City 69.9*** 59.2*** Test of difference between mobile and landline samples: Chi-square (d.f. 1) 25.0*** Table displays z-test of proportion of sample compared with Australian population *** p<.001; ** p<0.01; * p<

31 Table 4: Unweighted Estimates of Prevalence of Entrepreneurial Activity across Different Samples Percentage of Population (Confidence Interval) Landline Mobile Combined Sample Sample Sample Nascent Entrepreneurs Early Stage Business Owner Managers TEA: Total Early Stage Entrepreneurial Activity ( ) ( ) ( ) ( ) ( ) ( ) 5.4** 8.3** 6.9 ( ) ( ) ( ) Confidence intervals are two-side Table displays z-test of difference between mobile and landline samples. *** p<.001; ** p<0.01; * p<0.05; p<

32 Table 5: Weighted Estimates of Prevalence of Entrepreneurial Activity across Different Samples Percentage of Population (Confidence Interval) Landline Mobile Combined Sample Sample Sample Nascent Entrepreneurs Early Stage Business Owner Managers TEA: Total Early Stage Entrepreneurial Activity ( ( ) ( ) ( ( ) ( ) 5.2** 7.6** 6.6 ( ) ( ) ( ) Confidence intervals are two-side Table displays z-test of difference between mobile and landline samples. *** p<.001; ** p<0.01; * p<0.05; p<