AUSTRALIAN BROADBAND TELECOMMUNICATIONS RESEARCH PROJECT

Transcription

1 AUSTRALIAN BROADBAND TELECOMMUNICATIONS RESEARCH PROJECT Household Broadband Adoption Report Supplied to the Department of Communication, Information Technology & the Arts January 2004

2 Household Broadband Adoption Report FOREWORD This study was undertaken for the Department of Communications, Information Technology and the Arts by the Electronic Commerce Centre at the Australian National University. The project team consisted of: Professor Shirley Gregor (Project Leader) * Associate Professor Michael Martin (Project Leader) * Sigi Goode * Dr Steven Stern (primary responsibility for statistical analysis) *( Associate Professor John Rolfe ** * Faculty of Economics and Commerce, Australian National University, Canberra ACT 0200 Phone (02) ** Faculty of Business and Law, Central Queensland University, Rockhampton QLD 4702 Phone (07) Copyright notice and contact address to be inserted. 2

3 TABLE OF CONTENTS Page FOREWORD TABLE OF CONTENTS LIST OF ACRONYMS EXECUTIVE SUMMARY 1. INTRODUCTION 2. BACKGROUND 2.1 Broadband in Australia 2.2 Broadband outside Australia 2.3 Factors influencing adoption and use of broadband 3. METHODOLOGY 3.1 The data 3.2` Statistical methods 3.3 Research variables 4. THE BROADBAND PICTURE AUSTRALIA WIDE 5. RELATIVE IMPORTANCE OF FACTORS DETERMINING TAKEUP 6. AFFORDABIITY 7. USAGE 8. TECHNOLOGY COMPATIBILITY 8.1 Compatibility dimensions 8.2 Technophilia

4 8.3 Pay TV effect 9. ATTITUDES 10. DEMOGRAPHICS 11. CONCLUSIONS REFERENCES AND BIBLIOGRAPHY APPENDIX 1: TIMELINE OF SIGNIFICANT NETWORKING AND BROADBAND EVENTS IN AUSTRALIA APPENDIX 2: FACTORS FOUND TO INFLUENCE 63 BROADBAND ADOPTION IN PREVIOUS STUDIES APPENDIX 3: CLASSIFICATION TREES METHODOLOGY APPENIDX 4: BROADBAND TAKEUP RATES BY QUARTER, NATIONALLY AND REGIONAL APPENDIX 5: VARIABLES INCLUDED IN THE TREE 73 ANALYSIS APPENDIX 6: MAIN PURPOSES OF INTERNET USE

5 LIST OF ACRONYMS ABS ACCC ADSL DCITA DSL ISDN ISP Kbps Mbps OECD SDSL Australian Bureau of Statistics Australian Competition and Consumer Commission Asymmetric Digital Subscribe Line Department of Communications, Information Technology and the Arts Digital Subscriber Line Integrated Services Digital Network Internet Service Provider Kilobytes per second Millions of bytes per second Organisation for Economic Cooperation and Development Symmetric Digital Subscriber Line 5

6 EXECUTIVE SUMMARY Broadband communication technologies offer Australian households many opportunities, from greater access to education, health and government services, to entertainment and social connectivity. Australians appear, however, to be lagging behind other nations in adopting broadband. The Australian Competition and Consumer Commission estimated in October, 2003 that only 3% of Australians had broadband connections, compared with rates of 5% in the United Kingdom and 7% in the United States. The focus of this report is on the relative importance of factors affecting the takeup and usage by households in Australia of broadband technologies. An improved understanding of propensities to adopt broadband by households of different types can inform government policy and assist businesses designing strategies aimed at consumers. The study has used a large data set provided by a commercial market research company. The data covers the eight quarters in the period January 2001 to December Of the 23,093 records considered in the data set analysed for this report, 932 indicated a broadband home Internet connection giving a raw, unweighted take-up rate of 4.04%. Broadband services in this dataset were indicated by a home Internet connection of greater than 64 Kbps. The data set was weighted to provide an accurate picture of adoption Australia wide. An innovative statistical technique known as classification tree analysis was used for data mining of the sample studied, initially including 65 variables that could potentially help explain adoption. This technique has advantages over other methods such as regression analysis in that it can handle large data sets and numbers of explanatory variables and also offers a visual tool (a tree structure) that depicts complex interrelationships among variables. Though the classification tree analysis is exploratory, the majority of the variables found to be significant in explaining uptake were congruent with constructs found in the theory of innovation diffusion and in other studies, and the explanatory factors have been interpreted in terms of this prior work. Important factors Australia-wide indicating a household s propensity to adopt broadband, in approximate order of importance, were: Frequency of usage of the Internet (indicating needs and benefits of use); Location, Brisbane, Canberra, Melbourne, Perth, Sydney, and the non-metro areas of South Australia and the Northern Territory had higher use. This factor is likely to indicate, in part, the availability of services; Technophilia, a tendency to acquire technologies in general in the household; Pay TV subscriber, past or current, a further indication of technology compatibility and possibly also of the bundling of services in some areas (e.g. cable with Pay TV and broadband). The order of variables is stated as approximate, as the tree analysis shows the importance of factors relative to a subset of the sample; or in other words, the branch of the tree that is determined by the other important variables. 6

7 Other variables that are shown in the tree as important are: Times of main usage (both evening and other); Purpose of use (software download); Affordability indicated by main income in the household (greater than $120,000); Length of Internet usage. Some variables will not appear in the tree analysis because they are correlated with variables that do appear. Thus, while the variables selected are the best indicators of takeup, other variables remain important indicators of broadband takeup, and those relationships are also investigated in this report. Analyses were performed on the simple effects of attitudes and demographics, separate from the tree analysis; including education, household life-cycle, and number of children in the household. In general, the analyses raised few surprises and clear relationships emerged consistent with wellknown constructs in the theory of diffusion of innovations. Some interesting issues, however, are elucidated below. The study has produced two new insights that are important in describing broadband takeup in Australia that do not appear to be familiar from overseas studies and experiences. The first is that at this stage of the adoption cycle, it does not appear that cost is a primary driver of adoption. In contrast, the needs of and benefits to users is a primary driver. While the data set provided did not include direct data on pricing/cost of services, information relating to income and average amount spent provide reasonable proxies for cost. The main driver described in the tree analysis was frequency of use, suggesting that adopters in Australia are mainly interested in being able to access fast online services on demand, when they want them, because they are heavy users of the Internet. While cost-related variables do appear in the tree, they appear most relevant for users who do not access the Internet frequently. Thus, the picture painted is one where users sign up for services because they have a need for the type of fast, always-on service provided. Users with less need for this feature still sign on, but typically when they can afford to do so. In this context, cost remains an important factor in the adoption decision, but it is less important than the need/benefit driver. A second key finding of this study is that adoption appears strongly associated with a general predilection for technology (named technophilia) among adopters. This feature of adopters has not been prominent in other studies, but is broadly associated with the concept of technology compatibility and convergence of technologies. One explanation for this finding, consistent with the other variables selected as of primary importance, is that broadband services respond to individuals desires for services that do what they want, when they want, and form an integral part of the modern digital lifestyle that increasingly integrates information services into everyday lives. 7

8 1. INTRODUCTION Broadband network technologies are a significant advancement in the telecommunications industry. The opportunities for businesses and households provided by broadband are many and include increased productivity, avenues for education and learning, entertainment, greater personal convenience and increased social capital in communities. In Australia, however, the usage of broadband technologies by households and businesses lags behind other OECD countries. The reasons for the relatively low take-up rate of broadband in Australia are unclear, though cost and availability are considered to be prime concerns. The focus of this study is the use of broadband in Australian households. The primary aim of the study is to: Determine the relative importance of factors affecting take up and usage by households of broadband telecommunications services in Australia. The factors affecting broadband usage by households investigated included: Geographic location (region); Affordability (income level); Usage of the Internet and broadband (frequency, time and type); Technology compatibility ( technophilia, length of Internet use, Pay TV subscription); Attitudes to Internet technology (including security and privacy concerns); Household demographics. The relative effect of these factors was examined using a sophisticated statistical technique known as tree analysis, which is useful for mining information from large data sets. This technique provides results in a visual form (trees) that give insights into the relative effects of factors and aids in interpretation. The large data set examined covered all regions of Australia and was provided by a commercial market research organization. Broadband has been defined by the Broadband Advisory Group (BAG, 2003) as: The ability of a single access line or wireless satellite link, connected to a telecommunications network, to provide support for fast, always-on access to digital content, applications and a range of services, some of which can occur simultaneously. The first generation of Internet services used dial up modems and standard ISDN links to connect users to the Internet at narrowband speeds. The second broadband generation of Internet services depends on ADSL, cable modems and certain corporate and satellite data connections. This level is usually described as broadband if it provides always-on data services of 200 kilobits per second (Kbps) or more. In this study broadband was distinguished as greater than 64Kbps. 8

9 2. BACKGROUND 2.1 Broadband in Australia A number of reports on telecommunications and broadband issues in Australia have appeared, including the Estens enquiry report (Estens, 2002), the National Bandwidth Enquiry (DCITA, 1999), the Broadband Snapshot (DCITA, 2001), and Australia s Broadband Connectivity, a report from the Broadband Advisory Group (BAG, 2003). In the Broadband Snapshot report (2001) produced by the Department of Communications Information Technology and the Arts, demand drivers including price, availability, content and applications were proposed as the drivers of broadband connections (DCITA, 2001). This report found that with regard to price, customers desired more services from broadband connections rather than simply fast connections for the additional fees, and hence many still preferred the current Internet connection system. The paper suggested that more information is needed regarding price elasticity in this area to conclude whether price is a major demand driver in the adoption of broadband services (DCITA, 2001). The report Australia s Broadband Connectivity from the Broadband Advisory Group (BAG, 2003) recommended a National Broadband Strategy in cooperation with all levels of government and industry stakeholders. Further the government should consider initiatives and incentives in areas that are not commercially viable, but could potentially deliver significant economic, security and social benefits and encourage demand aggregation strategies. The areas of health, education and research were identified as foci for broadband initiatives. Take-up rates of broadband by businesses and households in Australia are monitored by the Australian Competition and Consumer Commission (ACCC). ). Table 1 shows statistics from the ACCC for recent time periods. The definition of broadband used by the ACCC is that stated in the BAG report (2003), and refers to services offering 200 Kbps or higher. Thus, the figures in Table 1 are not directly comparable with the data from this study, which uses a more inclusive definition of broadband. Table 1: ACCC statistics for broadband adoption (ACCC, June, 2003) Date Business Household Other Total 2001 July 16, ,100 2, , March 35, ,300 14, , June 57, ,500 27, , September 66, ,700 36, , December 70, ,700 59, , , March 75, ,600 83, , , June 84, , , , , Oct? Appendix 1 shows a timeline of events relating to Broadband in Australia. The ACCC report in October, 2003, showed that nearly 100,000 broadband connections were switched on in Australia in the June quarter, bringing the total number to 516,900 nation-wide. 9

10 Australia is lagging behind other nations in adopting broadband. An ACCC spokesman, commenting on the report, said the new data showed that only 3 per cent of Australians had broadband connections. The UK was running at about 5 and the US at 7. (Canberra Times, 13 Oct, 2003). 2.2 Broadband outside Australia The report Broadband Access for Business (OECD, 2002) examines the development of broadband access for business users in OECD countries. Though focussed on business, this report also has implications for household use of broadband with its recommendations concerning infrastructure. The report concludes that infrastructure competition continues to be the best way to develop broadband access. In many countries the level of competition is insufficient resulting in a negative impact on users. Countries with alternative infrastructure available to business users are developing broadband access much faster than in those markets where there is only one, or at best two platforms available to provide broadband access. (OECD, 2002, p. 4). Table 2 shows the takeup of broadband in a number of countries as shown in the OECD report (OECD, 2002). These figures show the Australian rate of 1.4% is below the averages for the OECD (3.9%) and the European Union (2.3%). The high takeup rate in Korea is particularly interesting. A report produced by Brunel University on the Korean situation (DTI Technology, 2002) states that Korea has the highest penetration of broadband in the world, with more than 60% of the households connected as of June The report lists six key factors that have contributed to this success: (i) geography and demographics, (ii) government vision, strategy and commitment, (iii) facilities based competition, (iv) the PC Bang phenomenon (online multi-player games), (v) price and (vi) clear user benefits. Table 2 Broadband subscribers in OECD countries Country Broadband subscribers per 100 inhabitants June 2002 Australia 1.4 Austria 4.2 Canada 10.3 Korea 19.2 New Zealand 1.1 United Kingdom 1.3 United States 5.8 OECD 3.9 EU Factors influencing adoption and use of broadband A number of prior studies have considered the influence of various factors on the adoption of broadband. Appendix 2 gives a list of these studies with the factors they have identified as influencing take-up. Broadly speaking, these influences can be grouped into a number of important categories which coincide with theory on the adoption of innovations (Rogers, 1995). Thus we find factors associated with: 10

11 Relative advantage, whether the technology yields advantage in terms of economic benefit, social prestige, convenience or satisfaction. This category is associated with concepts of costs, needs, benefits and risks. The cost of a technology is associated with affordability, which is indicated to some extent by household income level. Household demographics can also indicate specific needs or the desirability of technology use; for example, a household with children could view the use of broadband for entertainment (games) as especially desirable. Potential benefits are likely to be negated by perceptions of risk associated with the technology; that is, lack of trust in Internet security can overshadow needs for Internet use. Compatibility, the degree to which the technology is viewed as being compatible with existing values, past experiences and needs of potential adopters. This category is associated with concepts of prior experiences with similar technologies; for example, it may be indicated by the length of time the Internet has been used. Complexity, the degree to which a technology is perceived as being difficult to understand and use. This category is associated with demographic factors such as age, level of education, positive attitudes to computer use and so on. Trialability, the degree to which the technology can be experimented with on a limited basis. Thus, household users who have had the opportunity to use and gain understanding of broadband in their employment are expected to be more ready to adopt it at home. Observability or transparency of benefits, the degree to which the results of using the technology are visible. This category is associated with the way in which the technology benefits are presented to and perceived by potential users. The factors above can be considered as demand-side factors, as they encompass consumer needs, attitudes and readiness for broadband. In addition, there are a number of supply-side factors that affect uptake, including: Current broadband infrastructure environment, including availability and reliability of services, nature and cost of services available. The availability of security mechanisms and services is also important. Industry-specific pressures, in cases where household use is at least in part associated with engagement in some business-related activities; for example, telework, or research carried out from the home. If a household occupant is engaged in an industry such as graphic design, then industry pressures and the need to stay competitive will force consideration of broadband usage. Availability of content, whether required services or products are available. The presence of compelling content (referred to as killer apps ), which greatly stimulates demand is of interest. Education and support, in terms of information availability and education programs designed to stimulate uptake and reduce complexity and uncertainty of adoption and make benefits more transparent. Champions or case studies of success can increase the transparency of benefits. 11

12 3. METHODOLOGY 3.1 The data The statistical analysis of broadband adoption in Australia in this report was based on the Roy Morgan Telecommunications Monitor dataset provided by the Department of Communications, Information Technology and the Arts (DCITA). The investigation is exploratory in nature, using the data itself to drive model development. The dataset contained a large array of communication and media-oriented responses from nearly 100,000 participants interviewed over a four-year period. However, information directly concerning broadband adoption was collected only in the final two years, and only a subset of the respondents provided answers to the most relevant questions. As such, the analysis presented here is based on the 23,093 respondents for whom a valid response was recorded to the questions regarding the existence and type of home Internet connection. The total number of these respondents who indicated that they had a home connection with a speed greater than 64k was 932, giving a raw overall take-up rate across the two year period of 4.04%. The data were provided by DCITA under their agreement with Roy Morgan Research. A brief description of how the data set analysed for this report was extracted from the full data set follows. The initial data set consisted of records on 99,295 sampled households over 16 quarters, with responses provided to a large number of telecommunications-related questions along with a number of demographic variables. For the purposes of this study, only time periods during which questions related to broadband take-up were relevant, so records from January 1999 to December 2000 were removed, as during that time period no information was gathered from households as to whether they had broadband connections. This action reduced the size of the data set to 51,919 records. Further records were removed where there were missing values or can t say responses to the questions asking about Internet connections. 1 Of the 23,093 records ultimately considered in the data set analysed for this report, 932 indicated a broadband home Internet connection for a raw (unweighted) take-up rate of 4.04%. The data was collected and analysed across sixteen quarters (from January-March 2001 to October- December 2002) and a wide range of geographic and demographic areas. The dataset provided was weighted according to geographic region (11 categories: Sydney, Other New South Wales including the Australian Capital Territory, Melbourne, Other Victoria, Brisbane, Other Queensland, Adelaide, Other South Australia including the Northern Territory, Perth, Other Western Australia, and Tasmania) and overall household size (3 categories: 1-2 Persons, 3-4 Persons, and 5 or more Persons). The weights were provided by Roy Morgan Research as part of the data, and these weights were used throughout the analysis. Using this weighting scheme, the adjusted overall broadband adoption across all quarters was 4.18%. There are some characteristics of the data set which should be noted. The data is based on responses from one individual in the household. Some questions are based on self-reported attributes of this 1 This data set was reduced in size to records on the removal of households that provided no indication of whether the household had a home or work Internet connection (that is, for which the data set recorded a missing value for each of the ICONNC1, ICONNC2 and ICONNC3 [Internet CONNection] variables). Records were then removed for which respondents answered can t say to the question of what type of Internet connection existed in the household. The resultant data set contained records. Finally, records were removed for which the household indicated a home Internet connection (ICONNC1=1) but for which the respondent listed don t know as to the type of question and did not also list a broadband response (that is, TCONNC6=1 and TCONN4 not equal to 1 [TCONN reflects Type of CONNection]). The final data set for analysis therefore contained records, spread over 8 quarters. This number includes 3896 households that positively identified no Internet connection, but does not include the respondents for whom the ICONNCx (Internet CONNection) values are missing in the time period under consideration (January 2001 to December 2002). 12

13 individual, while others are based on reports of attributes of other individuals (for example, the main income earner) while still others concern the household as a whole. These distinctions are indicated as far as possible in the analysis that follows. In general, variables that represented the household as a whole, such as household income, were preferred for analysis rather than demographic information such as gender age and education level, which applied to the individual who responded to the survey. These individual responses cannot reliably give a representative picture of the household as a whole as the data set did not record which member of the household (including children) responded to the survey. 3.2 Statistical methods Classification and discrimination among sub-groups within a defined population are well-studied statistical problems. Traditional approaches are based on parametric models such as the multivariate normal and logistic regression. These methods attempt to relate the propensity for membership in various sub-groups among individuals to various covariate values. Typically, the fitting of such models requires heavy structural model assumptions that are difficult to verify in practice. Moreover, the models require a priori assumptions regarding the nature of the relationships between the covariates and sub-group membership. More modern, computer-intensive classification and discrimination techniques have been studied over the last decade and are now finding uses in a wide array of practical settings. These techniques, which include recursive partitioning classification trees, neural nets and so-called genetic algorithms, replace standard parametric model assumptions with computationally intensive searches, allowing classification model determination to be more directly data-driven. As such, the need for a priori assumptions regarding overall relationships between covariates and responses are dramatically lessened. For example, classification trees are invariant to covariate scale and can automatically incorporate potential covariate interaction. In addition, these modern techniques provide excellent end-product visualisation, allowing for a graphical interpretation of results, as well as generally simplified membership prediction rules. Traditional parametric classification and discrimination procedures are well outlined in many standard texts; see for example McCulloch and Nelder (1989) and Johnson and Wichern (1982). Recursive partitioning classification trees were introduced by Brieman et al (1984) and an excellent discussion of their current implementation is available in the recent monograph by Hastie, Friedman and Tibshirani (2001). The increased interpretability and flexibility of data mining techniques in general, and classification trees in particular, allow for greater ease of insight into many complex problems. Also, classification tree algorithms are readily adapted to handle the difficult issues of multiple response outcomes as well as observations with missing predictor values. The latter is an extremely common problem for survey-based data sets, and is not easily handled by the classical techniques, leading to sometimes substantial loss of information. A detailed description of classification tree methodology is provided as Appendix 3 to this report. Appendix 4 gives broadband takeup rates in numerical form Research variables This tree analysis method has the capacity to mine large data files and incorporate a large number of potential explanatory variables. Appendix 5 gives the full list of variables from the data set that was included in the analysis. All variables that potentially had a relationship with the factors identified as important in prior studies and from existing theory were included, leading to a total of 65 variables. As was expected, not all factors considered were found to be important in the tree analysis. The major factors identified in the analysis as affecting broadband usage by households investigated can be grouped under the construct headings of: 13

14 Location, geographic location of the household. This construct is in part an indicator of availability, as different infrastructure, services and pricing plans are available in different parts of Australia. Affordability, indicated by income level. This construct gives an indication of the cost of broadband relative to the amount of money available for expenditure in the household. Usage of the Internet and broadband, which represents both needs and benefits. This construct has three sub-constructs; o o o o Usage frequency, how often the Internet is used (at home or work); Usage home frequency, how often the Internet is used at home; Usage type, the purposes for which the Internet is used; Usage time, the period of the day in which the Internet is used. This feature is particularly interesting and suggests that the always-on feature of broadband connections allows adopters flexibility in managing their time and activities on the Internet. Technology compatibility, the degree to which Internet technology is compatible with other interests of occupants of the household. There are two sub-constructs here: o o o Length of Internet Usage, how much prior experience there is with a supporting technology; Technophilia, a tendency to acquire technologies in general. This construct in not one that is well-known in the literature, but it appears to be an emerging factor in discussions of the digital lifestyle. For example, the concept of convergence, where technologies become multipurpose (e.g. on-demand movies serviced over a cable/satellite connection which also provides broadband services; the Internet refrigerator that can sense when food needs to be purchased, and then the purchase is carried out over the Internet), is clearly related to the ownership of multiple technologies. In a sense, technophilia is a natural extension of the compatibility construct, reflecting the users experiences with particular technologies, but also building in users desires to further meet their needs/desires with further (possible compatible) technologies. Pay TV subscription, both past and current. This factor may be also an indicator of availability and bundling of services. For example, several service providers (e.g. TransACT cable service in ACT) offer a variety of services with attractive discounts for the bundling of services. Such bundling arrangements can even transcend traditional information areas one example is Actew/AGL providing discounts to users who bundle electricity services with telephone and Internet services. Pay TV subscription may also reflect aspirational or status-related feelings among consumers, and so may also be related to socio-economic status, income, and household life-cycle variables; Attitudes to Internet technology, including security fears, technology fear versus comfort with technology. 14

15 Demographics, including education level, number of children in household, number of persons in household. Some factors in the attitudes and demographics categories are considered in more detail in this report even though they did not prove to be significant in their own right in the tree analysis, in view of their general interest. These variables include education level and some of the attitudinal variables. It is likely that these variables did not appear in the tree analysis because of their correlation with other variables that were significant. The dependent variable in the study was formed from the answers to the following questions: Does the household have a home or Internet work connection? Type of Internet connection at home? Households that had both an Internet connection and indicated that their connection had a speed greater than 64Kbps were classified as Broadband adopters. Households that indicated that they either had no Internet connection or had an Internet connection with a speed no greater than 64Kbps were classified as non-broadband adopters. Households where don t know or missing values were given for either of the above questions were omitted from analysis. Table 3 shows the indicators for each of these constructs in the data set. 15

16 Table 3 Labels and indicators of main explanatory factors considered Construct Label Sub-constructs Indicator variables and range of values Location Geographic region (12 categories: Sydney, Other New South Wales, the Australian Capital Territory, Melbourne, Other Victoria, Brisbane, Other Queensland, Adelaide, Other South Australia including the Northern Territory, Perth, Other Western Australia, and Tasmania) Affordability Household income From less than $6,000 to greater than $120,000, in bands of $10,000. Main income Income of main income earner. From less than $6,000 to greater than $120,000, in bands of $10,000. Usage Usage frequency Classed as: (home or elsewhere) Heavy use the Internet more than once a day Moderate use the Internet between once a week to once a day Light use the Internet less than once a week Usage home Categories ranging from more than once a day to frequency less than once a year. Usage time Time periods most often use the Internet: Day Evening Other Various combinations of the above Usage type Main use of the Internet/broadband: Academic/business research (banking and insurance) Visiting web sites (information and entertainment, magazines, newspaper, radio, classifieds, buying/selling, auctions, software download) Shopping/paying bills Socializing/entertainment Promotion/publishing General information use Technology compatibility Length of Internet Usage Technophilia Subcategories of each of these main categories were available and used. How long the respondent has been using the Internet Technology Usage Index. Presence in household: +1 for: Digital TV or set top box +1 for: DVD Player +1 for: MP3 Player/Mini-disk player +1 for: Digital video camera +1 for: Digital camera valued over $ for: Fax +1 for: Any Games Console 16

17 Attitudes Social demographics Pay TV subscription Security/privacy fear Lack of security/privacy fear Technology comfort Technology discomfort Education level Children in household Persons in household Pay TV subscriber: Current Previous Never Concern with: Invasion of privacy Control of content for family Comfort with: Giving credit card details Grocery shopping. Comfort with: Learning new technology Control from computers Enjoy on-line chat Enjoy Discomfort with: Information super-highway Speed of change Intimidated by complexity Education level of respondent: Primary through to degree. No children at home, one, two, three or more children. Number of people aged 14+ in household 17

18 5. THE BROADBAND PICTURE AUSTRALIA WIDE The following figures show graphically the position regarding take-up in Australia geographically and over time. The information presented has both temporal and spatial components, and reflects a broad increase in takeup in Australia during the period under study. Figure 1 shows that, overall, national takeup has increased from 2.17% in January-March 2001 to 5.72% in October-December Between the October-December quarter 2001 and the January-March quarter 2002, there was a slight but statistically significant dip in takeup. This dip may reflect consumers reaction to changes to capped-download broadband offerings and reliability problems with ADSL networks that occurred in the latter half of Percentage Take up Jan Mar 01 Apr Jun 01 Jul Sep 01 Oct Dec 01 Jan Mar 02 Apr Jun 02 Jul Sep 02 Oct Dec 02 Figure 1: Overall takeup Australia-wide (weighted by 11 regions and household size) 18

19 0.65% 2.33% 2.20% 5.87% 3.19% 1.84% 1.49% 1.68% 6.30% 7.24% 4.91% 1.21% Figure 2: Geographical take-up rates (weighted by household size and quarters) Figure 2 presents average take-up rates by region, weighted by household size and quarter. Notable features are the relatively low take-up rates in regional Australia, almost certainly indicative of low availability in these areas, and the high take-up rates in major metropolitan areas serviced by multiple providers and multiple technologies. Figure 3 presents the same information using coloured areas to represent take-up rates in appropriate ranges. 19

20 >8% Take up 5 8% Take up 3 5% Take up 2 3% Take up 1 2% Take up <1% Take up Figure 3: Geographical take-up rates (weighted by household size and quarters) In this graphic, the dominance of major metropolitan areas is more obvious. The change in take-up rate through time is also of interest, and Figure 4 depicts takeup rates around Australia from the first quarter of 2001 through the last quarter of The graphic demonstrates that all areas of Australia have experienced growth in broadband takeup over the period of the study, but that the growth has been slower in regional areas (e.g. regional Western Australia and regional Queensland), erratic in some areas (NT/SA, reflecting relatively small samples in those areas), and rapid in major centres (Sydney, Melbourne, Canberra, Brisbane, Perth). It is expected that availability has been a major factor in this growth, and in influencing the differential rate of growth observed in the graphic. 20

21 January March, 2001 April June, 2001 July September, 2001 October December, 2001 January March, 2002 April June, 2002 July September, 2002 October December, 2002 Figure 4: Broadband take-up depicted by region over 8 quarters from the beginning of 2001 to the end of 2002 (The colour legend is as for Figure 3) 21

22 5. RELATIVE IMPORTANCE OF FACTORS DETERMINING TAKEUP The most relevant predictors, as chosen by the cross-validation criterion for tree size determination, are: frequency of Internet usage, geographic region, household income, income of main earner, length of Internet use, Pay TV subscription, time of day for main Internet usage. One useful way to interpret the output of a tree is to construct groups of nodes where predicted usage is high, to investigate defining characteristics of likely users versus non-users. In the current tree, if we group together all terminal nodes with an observed usage larger than the overall average prevalence (4.18%), the group of adopters is characterised as: (a) heavy Internet users (those who use the Internet more than once per day) and either: (i) live in the ACT, Sydney, Melbourne, Brisbane, Perth; or, (ii) are technophiles (Technology Uptake Index exceeding 1). (b) Those who use the Internet no more than once per day (light/moderate usage) and either: (i) are serious technophiles (Technology Uptake Index exceeding 2) (ii) live in the ACT, Sydney, Melbourne, or Brisbane, and currently subscribe to a Pay TV service. In general, we see that broadband take-up appears to be most prominent among those in the major population centres who either use the Internet very frequently or are technophiles. Notably, there seems to be a relationship between broadband take-up and pay television subscription, particularly among light Internet users, those who do not use the Internet more than once a day. This phenomenon may reflect a bundling of services by media providers or a general propensity for consumption of all media services. Finally, if we increase our threshold for inclusion into the group classified as adopters to, say 9%, then this removes all households outside of the major capital cities and incorporates purpose of usage (e.g. software download), income, time of main usage, and pay TV patterns. In particular, this group of adopters typically uses the Internet for software download as well as other purposes, or has an income exceeding $120,000, or uses the Internet more than once a day and has used the Internet for over 3 years and currently subscribes to a Pay TV service. In interpreting these results, it is important to remember that the location variable may well be a partial surrogate for availability, not only of potential for household connections but of basic infrastructure as well. Moreover, as many of the explanatory variables are likely to be inter-related, the absence of a predictor from the final tree does not necessarily mean it is unrelated to broadband take-up, but could simply be subsumed in the relationship between broadband adoption and other variables. For instance, there appears to be no connection between broadband take-up and mobile phone ownership or type of dwelling. However, these variables may have an effect which is already captured by the variables included in the tree model. These effects are explored more fully by examining graphs of takeup by category of individual variables. 22

23 % Take Up 4.18 Internet Usage: Light/Moderate Internet Usage: Heavy Tech Index <= 2 Tech Index > 2 Other Regions Brisbane, Canberra, Melbourne, Perth, Sydney, Rest of SA + NT Other Regions Not a Current PayTV Subscriber Brisbane, Canberra Melbourne, Sydney Tech Index <= 1 Current PayTV Subscriber Tech Index > 1 Main Internet Uses: Not Software Download Tech Index <= Main Internet Uses: Software Download Never a PayTV Subscriber Tech Index > Past or Current PayTV Subscriber Main Income < $ Times of Main Income >= $ Main Usage: Not Evening or Not Other Times of Main Usage: Both Evening & Other Home Internet Usage <= Once/day Home Internet Usage > Once/day Used Internet <= 3 years Used Internet > 3 years Not a Current PayTV Subscriber Current PayTV Subscriber Figure 5: Tree model for broadband take-up Australia-wide 23

24 Exhibit 1 Profile of typical adopters Either a) heavy Internet users (those who use the Internet more than once per day) and either: Or, i.) live in the ACT, Sydney, Melbourne, Brisbane, Perth; or, ii.) are technophiles (Technology Uptake Index exceeding 1) b) Those who use the Internet no more than once per day (light/moderate usage) and either: i.) are serious technophiles (Technology Uptake Index exceeding 2) ii.) live in the ACT, Sydney, Melbourne, or Brisbane, and currently subscribe to a Pay TV service. Separate classification trees were also created for users residing in each of Sydney or Melbourne, respectively. Trees for other major centres were not constructed because the amount of data available in each centre was considered insufficient for a reliable analysis. The goal of this analysis was to assess whether adopters characteristics differed between the major population centres. Broadly speaking, within the major centres, the characteristics of adopters were similar to those nation-wide, with a few relatively minor but interesting differences. For instance, for both Sydney and Melbourne, as nationwide, heavy Internet use and high Technology Uptake Index ( technophile ) were associated with broadband takeup. Also, in both centres, among light users, income was a significant factor influencing takeup. In the case of Sydney, however, among light Internet users with low Technology Uptake Index, interest in gaming was a factor influencing broadband takeup, while for Melbourne, among light Internet users with low incomes, interest in buying and selling over the Internet was a key factor influencing takeup. Overall, broadband take-up in Sydney slightly exceeded take-up in Melbourne (7.24% versus 6.48%), both well above the national rate. 24

25 %Take up 7.24 Internet Usage: Light/Moderate Internet Usage: Heavy Technology Usage Index <= 2 Technology Usage Index > 2 Technology Usage Index <= 1 Technology Usage Index > 1 Main Uses: Don t Include Gaming 3.40 Main Income < $110k Main Internet Uses: Include Gaming Main Income >= $110k Used Internet <= 5 years 9.02 Used Internet > 5 years Main Sites: Not Entertainment or Related Info Main Sites: Entertainment and Related Info Persons in Household Aged 14+ <= 3 Persons in Household Aged 14+ > 3 Main Uses: Don t Include Contacting Others Main Uses: Include Contacting Others Only Moderate Interest Comm. w/ Others Worldwide Quite Interested in Comm. w/ Others Worldwide # Private Phone Lines <= 2 # Private Phone Lines > Figure 6: Tree model for broadband takeup in Sydney 25

26 % Take up 6.48 Internet Usage: Light/Moderate Internet Usage: Heavy Household Income < $80k Technology Usage Index <= 1 Household Income >= $80k Technology Usage Index > Main Internet Sites: Not e Commerce Main Internet Sites: Software Main Internet Download Sites: e Commerce Main Internet Sites: Not Software Download Home Internet Usage < Once/Day Home Internet Usage >= Once/Day # Persons in Household <= 3 # Persons in Household > 3 Times of Main Usage: Daytime Times of Main Usage: Evening Main Internet Sites: e Commerce and Classifieds Main Internet Sites: Not e Commerce or Classifieds Home Internet Usage < Once/Day Home Internet Usage >= Once/Day Technology Usage Index <= 4 Technology Usage Index > Figure 7: Tree model for broadband takeup in Melbourne Differences between the Sydney and Melbourne trees are minor, and the similarity of these trees with the national tree is unsurprising given the dominance of Sydney and Melbourne in terms of population in general and takeup of broadband in particular. Sydney, Melbourne, and other major centres such as Brisbane, Canberra and Adelaide have available not only multiple providers of broadband services, but also multiple technologies for delivering broadband, supplementing ADSL and Satellite services with the availability of cable Internet services. Figure 8 shows the breakdown by type of broadband service for the regions considered in this study. Note the predominance of ADSL services in regional Australia compared with metropolitan areas, where both cable and ADSL services are available. Note also that in major metropolitan areas, the second largest provider (Optus) is available, while Telstra is dominant throughout regional Australia. 26

27 Sydney Telstra (cable) Telstra (ADSL Melbourne Telstra (cable) Telstra (ADSL) Telstra (cable) Brisbane Telstra (ADSL Optus Telstra (cable) Adelaide Telstra (ADSL) Optus N/A Other Optus N/A Other Optus Other N/A Other N/A Perth Canberra Other NSW Other WA Telstra (ADSL) Telstra (ADSL) Optus Telstra (cable) Telstra (ADSL) Telstra (ADSL) Telstra (cable) N/A N/A N/A Telstra (cable) Other Other Other Other Telstra (cable) Optus Other VIC Telstra (ADSL) N/A Telstra (cable) Other QLD Telstra (ADSL) N/A Other SA + NT Telstra (ADSL) N/A Tasmania Telstra (ADSL) N/A Other Other Other Other Figure 8: Service providers for those with high-speed connections by region (weighted by household size and quarter) 27

28 6. AFFORDABILITY Initial analysis showed that household income and income of main earner were key factors influencing takeup. This finding indicates that cost may be a significant factor, as higher costs can be more easily afforded in high income households. Figure 9 shows clearly that, ignoring other factors, income is clearly related to takeup in this study, with takeup rates soaring into double-digits for incomes exceeding $120,000. Incomes above $70,000 were associated with takeup rates above the national average, and takeup for incomes below about $40,000 was below half the national average. This finding suggests that cost is a factor inhibiting takeup, particularly for low-income earners. Of course, income is an imperfect proxy for cost as it also incorporates a number of other factors. Percentage Take up < Household Income (in $1000 s) Figure 9: Percentage take-up by Household income (1/ /2002) (weighted by 11 geographic regions, household size and quarter) Another way to explore cost issues related to takeup is to consider expenditure profiles for users of particular technologies; see Figure 10. The shape of this graphic is very illuminating: the vast bulk of broadband users spend $80 to $100 or more on the Internet per month, while both ISDN users and others (predominantly modem users) spend around $25 to $35 per month. Broadband is at least double, and up to three times as expensive in terms of typical users monthly expenditures than 28

29 narrowband technologies. Note that these statistics arise not from sampling the actual costs of services as advertised by providers, but from sampling users estimates of their expenditure. Nevertheless, the message seems very clear: broadband users spend considerably more, on average, than narrowband users, and it seems natural, therefore, that higher-income households should be associated with higher take-up rates. 1 Percentage of Households Broadband Users ISDN Users Other Users Monthly Expenditure Figure 10: Monthly Internet expenditure profile for home Internet users (weighted by 11 geographic regions, household size and quarter) 29

30 7. USAGE Usage is considered under the headings frequency of home usage and type of main usage. Frequency of usage was the most important factor identified by the tree analysis as being related to broadband takeup. Those users classed as heavy users (accessing more than once a day) were more than twice as likely to have broadband as those classed as light or moderate users (less than once a day). Figure 11 clearly shows the dramatic increase in takeup rates for those who access the Internet more than once a day. From a technology standpoint, the always on feature of broadband Internet allows users flexibility to be active on the Internet whenever they want rather than at set times at which they would manually make a connection. A typical modem user might partition their times and activities on the Internet so that they could use a single connection to achieve all their tasks for the day or week, while broadband users have the flexibility to participate in Internet activities at any time without incurring the cost of making a separate connection. Such convenience is consistent with the idea that broadband provides an additional benefit to consumers beyond increased speed. Percentage Take up Never <1/Year 1/Year 2+/Year 1/Month 2+/Month 1/Week 2+Week 1/Day >1/Day Frequency of Home Internet Usage Figure 11: Percentage takeup among home Internet users with various frequencies of home usage (1/ /2002) (weighted 11 geographic regions, household size and quarter) The usage profiles for broadband and other users (Figure 12) also reveals major differences in the frequency with which broadband users access the Internet. A typical broadband adopter accesses the Internet over 600 times per year, while narrowband users typically access the Internet many fewer times per year about a third as often. This phenomenon may be a consequence of the fact that narrowband users typically have to make a manual connection each time they access the Internet, 30