|Page tools: Print Page|
HOUSEHOLD ADOPTION OF DIGITAL TECHNOLOGIES
There is strong private and public sector interest in the diffusion of consumer technologies. Forecasts are often attempted - and are often wrong.
In the case of networked, online services, forecasts are particularly difficult, not least because of the dynamic and complex relationship between supply and demand forces which applies to all services delivered via a network infrastructure (see BTCE, 1995).
In the private sector, attempts at forecasting generally focus on identifying the socio-demographic characteristics of households which are most likely to enter the market in the short term. Public sector attention also encompasses that group least likely to be able to access commercially provided digital technologies - a group sometimes described as an information underclass.
In response to concerns about development of such a group, the Bureau of Transport and Communications Economics (BTCE) initiated a project in 1996 called Access to Information and Communications Services. This project aims to develop an analytical framework for examining the rationale and costs of public policies to minimise barriers to accessing online services from home. The central research questions are:
The analysis presented here of ABS data on digital technologies in the home seeks to address the first of these questions, and thereby to demonstrate the value of linking responses to several innovative ABS survey questions by means of a simplified, diagrammatic form of multivariate analysis called a tree diagram.
The ABS survey
The first ABS household survey on the use of information technology (IT) in the home was carried out in 1994. The results indicated marked differences in adoption of computers between households of different socio-demographic types. For example, at that time 47% of households in the top income quintile had a computer, compared with only 5% of those in the lowest quintile (ABS, unpublished data).
In response to strong private and public sector interest in its data, the ABS then embarked on an expanded household survey program. At the end of a four-stage survey cycle the 1996 data set will consist of 12,000 observations. The first results, from the February 1996 sample of 3,000 households, were published in September (ABS, 1996). This article is based on analysis of results from a combined data set consisting of the February and May 1996 survey results.
As a first step in the analysis presented here, the estimated 6.6 million Australian households were classified into one of four groups, on the basis of the presence or absence of the three technologies currently required to access digital services from home:
The relative sizes of these four groups are shown in table S3.1, which highlights the fact that the majority of Australian households are not yet 'being digital' - to use the phrase coined by one commentator on the information economy (Negroponte, 1995).
Indeed, the February/May 1996 data set indicates that the nearly 70% of all households lacked at least two of the three prerequisites for accessing digital data from home, namely computers and modems.
The February/May 1996 results show that the proportion of households which had already become digital was 30.3%, compared with 23% two years earlier. Are such increases, of the order of 10-15% per year, likely to be sustained? What proportion of Australian households might be expected to be digital in two years time?
To answer these questions, it is tempting to try to look into the future using straight line projections. For example, if the changes observed over the last two years were projected into the next two years, we would see:
However the history of diffusion of many new consumer technologies suggests that straight-line projections are unlikely to be very good predictors. This is mainly because not all products achieve the rapid growth associated with mass market take-up, so that the diffusion curve only acquires its familiar S-shape in the case of stable and successful products or services, such as the VCR. The availability of only two data points (for 1994 and 1996) means that it is not yet possible to identify changes in the rate of growth in adoption, and therefore to identify the current stage of market diffusion of digital technologies. Without this information, estimates of future demand can only be carried out through comparison with analogous products (see BTCE 1994).
With such caveats in mind, the projection frame proposed for this BTCE analysis is deliberately short (i.e. 1996-98). The purpose of the projection is not to generate forecasts as such - an assessment of market developments likely to influence take-up rates is beyond the scope of the analysis - but rather:
The pool from which new computer-using homes will be drawn is the four and half million or so households which the February/May data set indicates did not use a computer (i.e. 70% of 6.6 million households).
Growth scenarios have been based on links between responses by these non-digital households to three ABS survey questions which relate to:
These questions are each briefly described in the following section.
Intention to spend
In these surveys the ABS asked whether anyone in the household planned to purchase, upgrade or replace any of their computer equipment, including software, within the next 12 months or two years. The predictive power of responses to this question from the February/May 1996 surveys can be gauged by comparing the actual penetration rate in 1996 with the rate that would have been achieved if non-computer households which in February 1994 reported plans to purchase actually acquired a computer. The actual penetration rate achieved in February 1996 was 29.6% and the rate that would have occurred if the plans to purchase within the next 12 months from February 1994 were fully realised were 31.3%. By contrast, spending intentions at February 1994 with a two year period, rather than a one year period, significantly overstated actual penetration, since they yield a predicted rate of 37.2%.
Comparisons of 1994 and 1996 data sets suggest therefore that householders predict more accurately their actual purchase of computer equipment over the shorter of the two time frames.
When the combined February/May data set is considered in a similar way, the 12 month spending intentions question yields a penetration of 31.6%, compared with the actual penetration rate of 30.3%.
Reasons for not having computer and/or modem
The 1996 ABS survey sought, for the first time, information about reasons for householders' decisions not to acquire computers or modems. Respondents in households without a computer (or in the case of modems, with a computer but no modem) were asked to nominate one main reason for non-adoption from the following list:
These stated reasons offer some clues into factors which would need to change for households to acquire either a computer or a modem. The BTCE's analysis of non-computer using households involves grouping together households for which there is a stated lack of interest, or perceived lack of use for a computer, or a lack of skills, and distinguishing this group from those which state that cost, or access elsewhere (e.g. in the workplace) is the reason for not having a home computer. This division is based on the following two assumptions.
First, households for whom a lack of interest in, or perceived lack of use for, computers is the main barrier must be viewed as much less likely to purchase computers and/or modems even if costs fall.
Second, households for whom cost is the main barrier to equipment acquisition may be assumed to be willing to purchase computers or modems in principle, but to be subject to a binding financial constraint. Such constraints are likely to weaken as the costs of computing equipment fall in real terms, standards become established and a greater emphasis on the residential market by suppliers makes purchasing easier (and less risky) for consumers.
On the basis of these two assumptions, responses to the question about stated reasons for lack of past purchases are used in the analysis as an additional filter on the division of the non-digital population into those who intend to spend and those who do not.
Interest in online services
Information was also sought for the first time in the 1996 survey on interest in a range of online services. All households, irrespective of whether or not they used a computer, were asked whether they would consider using a television or computer for home-based shopping, banking, gambling and/or staying in touch with people or finding things out via electronic mail (e-mail). They were asked to ignore cost considerations for the purpose of the exercise.
The ABS survey aimed to avoid associations with particular technology platforms by asking respondents to consider whether they would be interested in the services delivered by means of a television or computer.
The various permutations of responses to the three questions referred to above can be examined systematically using tree analysis.
In this analysis, the more important the attribute, the closer it is to the top of the diagram (see Aitkin 1982 for a discussion of an Australian application of this form of analysis, which is useful for depicting complex relationships between socio-demographic attributes which are themselves strongly co-correlated. See Sonquist, Baker and Morgan 1971 for the methodology.) Therefore, since intention to spend already has demonstrated predictive power, households were first divided using responses to this question as the first discriminant.
The best ordering of the other two ABS questions - both of which were asked for the first time in 1996 - could not be determined without analysis to test the extent to which binary division of each group, using the response to a particular survey question, reduced variance in the sample. However, such statistical analysis is not readily carried out without access to data in electronic format. Therefore, reason for non-adoption was ranked as the second discriminant, on the basis that these responses are based on past, actual behaviour - in contrast to interest in online services, which is necessarily speculative about future interest in services which are likely to be unfamiliar to most households.
The next section presents the results of the tree analysis and is followed by a description of how the results may be used to assess the likelihood of different groups of households acquiring computers in the short term.
In diagram S3.2, a classification of non-computer using households is depicted, in which they are grouped on the basis of spending intentions, reason for not having a computer at home and interest in online services. At each branch of the tree, the number of households and the proportional split between the two branches at the same horizontal level is shown.
Figure S3.2 shows how the 70% of households which the February/May data set indicates did not have a computer at home responded to the three ABS questions considered here. For example, working down the left hand side of the tree:
In general, a stated intention to spend on computers seems to be linked with a response which cites cost or access elsewhere as the reason for not having acquired a computer to date: households with spending intentions are more than twice as likely to give this reason than those with no spending intentions (i.e. 64% compared with 29%).
However, the nature of links between future spending intentions and reasons for past non-adoption on the one hand and interest in online services on the other is not always so clear cut (and will be the topic of further BTCE analysis). For example, Cell B contains households which were interested in at least one online service - despite lack of computer expenditure plans and despite citing lack of interest as the main reason for not having acquired a computer. This group of around 0.7 million households therefore looks likely to include households which might seek to access services through a TV set-top box combination rather than through a PC/modem.
S3.2 STATED PAST AND FUTURE MARKET BEHAVIOUR OF NON-COMPUTER USING HOUSEHOLDS
Probability of acquiring a computer
Table S3.3 summarises a systematic allocation of each cell in daigram S3.2 into groups with descending likelihood of acquiring a computer over the next two years. The percentages shown in the final column highlight the relatively small size of those groups which have been assigned the highest probability of acquiring computers (eg 3.5% in Cell H, compared with 30.9% in Cell A).
S3.3 SCENARIOS FOR GROWTH IN COMPUTER-USING HOUSEHOLDS
In developing growth scenarios, the existing population of home computer users (nearly two million households) is used as a lower bound on the likely penetration in 1998. Scenarios of higher adoption are then derived by adding in groups which have different combinations of spending intentions, reasons for non-adoption and levels of interest in online services, as summarised in table S3.4. Households
most likely to acquire a computer are assumed to be those reporting plans to spend on computer equipment within the next 12 months; table S3.4 shows that over half a million households were in this category.
S3.4 POTENTIAL FOR GROWTH IN HOUSEHOLD COMPUTER USE - 1996-98
Source: BTCE estimates based on unpublished ABS data.
If all these non-computer-using households currently planning to make computer expenditures within 12 months were to acquire a computer by 1998, the household computer penetration rate would rise from 30.3% to 38.1% (compared with an estimate of 40.1% from straight line projection). Higher penetration would imply the addition of households with no current plans for computer expenditures.
In this analytical framework, those least likely to acquire a computer are those with no known computer spending intentions, which report lack of interest in computers as the reason for not having one at home and which are currently uninterested in using online services. More than two million households satisfied these criteria (i.e. around 30% of all Australian households).
To reiterate, the reason why this latter (very large) group has not acquired and does not intend to acquire digital technology is not cost-based. Rather it is based on lack of interest in both computers and the online services which they deliver to the home. This suggests that a realistic upper bound on household penetration of digital technologies is 70%, in the medium term at least (although of course, the least likely group might become willing to acquire digital technologies if, or when, a large proportion of everyday transactions like bill paying becomes available significantly more cheaply or conveniently online).
If the cost constraint were to be reduced - as is possible through economies of scale as computers achieve more widespread adoption - the scenario analysis suggests that penetration could increase by a further 10% or so to around 49%.
In the short term, a scenario in which a majority rather than a minority of Australian households became digital could only take place if price reductions were combined with more compelling applications than those described to survey respondents, or alternatively higher awareness of their relevance.
In the longer term, increasing IT literacy within Australia will also affect the degree of penetration of computers into homes.
The next step in the analysis is to examine the socio-demographic characteristics of the half million or so households identified as most likely to become digital, and compare these characteristics with those of the two million or so households least likely to do so.
Future diffusion pathways
As noted above, the two projection methods discussed (i.e. straight-line projections and those based on the tree analysis) generate very similar results at the aggregate level, (40.1% and 38.1%, respectively). However, when the most likely and least likely groups identified using the tree analysis are examined more closely, it is clear that different types of households have very different propensities to become digital.
Table S3.5 shows, for example, that households in the lowest income quintile constituted only 6.2% of the most likely group, but were 34.9% of the least likely group. The table also shows how these two groups are distributed between households of different structure: clearly, those with children are much more strongly concentrated in the most likely group (they make up almost half of the group), while households which included people over the age of 60 years are poorly represented (less than 3%), but make up more than half of the least likely group. Households in non-metropolitan areas and those without a home business are both relatively over-represented in the least likely group.
S3.5 CHARACTERISTICS OF HOUSEHOLDS MOST AND LEAST LIKELY TO ADOPT COMPUTERS BETWEEN 1996 AND 1998
(b) Households where a computer was not used at home at Feb/May 1996; which had no plans for computer or computer-related expenditure; which claimed that lack of interest in computers, lack of use for a computer, or lack of skills to use one were the reasons for the household not having a computer; and which expressed no interest inonline services.
(c) Proportion of households most/least likely to acquire a computer originating from each category.
(d) Does not add to 100% because excludes households not classified to a quintile (no income recorded); see St Clair et al (1996) for income ranges.
(e) See St Clair et al (1996) for composition of each household structure grouping.
Source: BTCE estimates, based on unpublished ABS data.
What if all the households in the most likely group were to acquire a computer over the next two years? What would the penetration rates be among the groups listed in table S3.5?
Table S3.6 shows that by 1998 penetration rates could exceed 50% among upper middle and high income households, at a time when those in the lowest income quintile could still have penetration rates closer to 10%. Similarly, more than 50% of households consisting of couples with children could have a computer by 1998, by which time less than 10% of households made up of older people could be similarly equipped.
When compared with similar disaggregations for the February/May 1996 data, the estimates in table S3.6 of the distribution of computer-using households in 1998 suggest a continuing steady increase in takeup among lower-middle and middle income households. This would result in a slight narrowing of disparities along income lines (i.e. from a 6.4 fold disparity between penetration in the lowest and highest income groups, to a 5.9 fold disparity), as takeup among these groups increases faster than among high income households - albeit from a lower base. This would be inconsistent with findings in the United States where such disparities are reported to be widening (RAND 1995).
(b) The most likely PC penetration in households by 1998 is formed by summing the groups which had a PC at February/May 1996 and the group most likely to acquire a PC by 1998.
(c) Does not add to 100% because excludes households not classified to a quintile (no income recorded) - see St Clair et al (1996) for income ranges.
Source: BTCE estimates, based on unpublished ABS data.
Propensity to acquire modems
A tree analysis can also be used to develop scenarios for the adoption of networked computers, i.e. to identify the groups least and most likely to become digital connected.
To use online services, households currently need both a computer and a modem. Households with a computer and modem are not necessarily users of online services; for some households, this equipment may be used to remotely access work- or education-related databases. However, for the purposes of this analysis, the terms modem owner and online service user are used interchangeably. Almost half a million households already own both equipment items. Common sense suggests that the one and a half million or so households which currently use a computer at home, but which do not have a modem, are prime candidates for acquiring a modem.
In a similar analysis to that used to identify the households most likely to adopt computers, a tree analysis can be used to develop scenarios for the size and composition of the digital connected group in two years time.
Households most likely to acquire a modem were assumed to be those which currently have a computer, which plan to spend on computers or computer-related equipment within the next 12 months and which are interested in online services. This group consisted of around 365,000 households. Adding these to the households which already have a computer and modem as at February/May 1996 would give an aggregate penetration rate of 12.9%, with a socio-demographic breakdown as shown in table S3.7.
(b) The most likely modem penetration in households by February 1998 is formed by summing the group which had a modem at February/May 1996 and the group most likely to acquire a modem by 1998.
(c) Does not add to 100% because excludes households not classified to a quintile (no income recorded); see St Clair et al (1996) for income ranges.
Source: BTCE estimates, based on unpublished ABS data.
This table suggests that the digital connected group is likely to continue to consist predominantly of higher income households, those with children, those in metropolitan areas, and those with a home business - in the short term at least. Older and low income households are noticeably under-represented, confirming the patterns observed for computers themselves.
However, tree analysis also suggests that there is another group, of similar size to the most likely group discussed above, which also might be considered likely to start using online services in the short term. This group consists of the 344,000 or so households which do not currently have a computer, but which are planning computer expenditures and which are interested in at least one online service (see Cells F and H in diagram S3.2). Given that many computers now have modems installed as standard equipment, the leap from telephone connected to digital connected is not unlikely for members of this group.
The final column in table S3.8 shows what the socio-demographic characteristics of the modem-owning population in 1998 would be if the adoption scenario shown in table S3.7 were to be augmented by addition of this next most likely group of households. In this scenario, aggregate penetration would be 18.1%, rising to more than 25% among upper middle and high income households and those with children.
The second and fourth columns in table S3.8 show how the households most likely and next most likely to adopt modems are distributed socio-demographically: it suggests that the next most likely group is quite similar in profile to the most likely group, albeit with a flatter income distribution, and a stronger representation of non-metropolitan households and households without home business.
Thus, if the next most likely group were to adopt modems, this, would be consistent with diffusion of digital connectivity into the wider community. Once again, as with computers, this would be the result of faster rates of growth among later adopters, albeit from a lower base.
For example, table S3.9 shows that in 1994, metropolitan households were 4.1 times more likely to be digital connected than those in non-metropolitan areas. By 1996, this disparity was 2.2, and would fall to 2.1 under the more conservative most likely scenario, and to 1.9 if households which currently do not have computers, but which seem likely to acquire a bundled computer and modem, were to do so by 1998.
The analysis presented here suggests that the proportion of households which will have computers could rise to around 40% by 1998. Penetration could rise to 60% in high income households and those with a home business. Higher levels would be achievable if lower prices were to make purchase more attractive for a group of around 10% of households who currently cite cost as their main reason for non-acquisition.
The proportion of households equipped to use online services could increase to levels around 13% within the next two years under the more conservative most likely scenario illustrated in table S3.7. Higher levels, about 18% in aggregate terms and as high as 37% in high income households, are conceivable if online services became sufficiently compelling for people to acquire a bundled computer and modem, as explored in the digital connected scenario illustrated in table S3.8.
Penetration rates higher than this would require households currently not using computers at home to acquire modems, or alternatively for the addition of households not presently planning any computer-related outlays in the next 12 months.
Under each of these scenarios, however, digital connectivity seems likely to remain very low among households with older persons, and to a lesser extent among single parent households. Households outside metropolitan centres look likely to continue to have a lower level of digital connectivity than their capital city counterparts, despite faster rates of growth in adoption.
The methodology outlined here offers a systematic means of examining the characteristics of households which are not only 'have nots' at present but which are currently the least likely to become 'haves' --- even when prices fall.
However, there are several limitations to the approach used.
First, the analysis focused only on barriers related to the equipment needed in the home in order to access digital services via networks. The influence of differing telecommunications network capacity could not be considered in this framework. Offering respondents two different cost options as their reasons for not having a modem would assist in considering this issue, that is if they were able to distinguish between the capital costs of acquiring a modem and telephone access charges (which are higher for people who have to use STD calls to their Internet service provider, and for those whose network infrastructure is poor relative to that available in metropolitan areas).
Second, the 'why not' questions were presented to respondents in the form of a predefined list of 'main reasons', as identified through open-ended questioning in pilot surveys. While this approach facilitates large-scale data collection and analysis, there is always some risk that important factors might be missed. In this context, recent qualitative Belgian research which sought multiple reasons for non-adoption of a range of technologies is of interest: the results suggest that cost is a relatively minor reason for non-adoption of computers, and that instead responses broadly classifiable as 'no need for one' account for most non-adoption (Punie, 1996). However, the extent to which these findings might be applicable in an Australian context is unknown.
Third, the predictive power of the ABS survey questions on past and future market behaviour remains to be fully tested. The only question with a track record is the one which relates to intention to spend on computers and computing equipment within a year --- and even here the predictions are accurate only when the time period between the stated intention to spend and actual expenditure is extended to two years. More work needs to be done in this area, not least in establishing whether there are significant differences in the accuracy of the predictions among groups with different socio-demographic characteristics.
Fourth, the results of the relatively simple multivariate analysis presented here does not disentangle significant correlations between, for example, income and geographic location. Other forms of analysis would be required to do this - a potentially fruitful area for future collaboration between the ABS and the BTCE.
These methodological issues aside, it is important to note that the analysis presented here cannot - and does not seek to - provide answers to key public policy questions like what should be done and why. The results do, however, highlight gaps in our present understanding of the impact of online services. For example, it is clear that researchers currently have little information on which to assess the influence of digital technologies on the ability of individuals to participate in the economic and social life of their communities.
Thus, while the analysis presented here suggests that households consisting of low income and older people are concentrated in the group which is least likely to acquire a computer or a modem in the short term, what is still unclear is the extent to which this lack may translate into any constraint on the ability of such householders to participate in society and the economy.
What would help would be a way of linking ABS data on household adoption of digital technologies with other ABS data on labour market status, educational outcomes, and the information intensity of industries --- in aggregate or, more usefully, at the local or regional level.
Analysis of such links could lay the foundations for improved assessments of the need for government intervention to influence the availability or takeup of digital technologies at the household level.
Aitkin, D. 1982, Stability and change in Australian politics, Australian National University Press, Canberra.
Australian Bureau of Statistics 1996, Household Use of Information Technology (8128.0), Canberra.
Bureau of Transport and Communication Economics 1995, Communications Futures Project Final Report, Report no. 89, AGPS, Canberra.
Negroponte, N. 1995, Being Digital, Hodder and Stoughton, Sydney.
Punie, Y. 1996, 'Rejection of ICTs in Flemish households: the why-not question', paper for EMTEL meeting on Media and Information Technology: Regulation, Markets and Everyday Life, Brugges, Belgium, 8-9 November 1996.
RAND 1995, Universal Access to E-mail: feasibility and societal implications, RAND, Santa Monica.
Sonquist, J.A., Baker, E.L., & Morgan, J.N. 1971, Searching for Structure, University of Michigan, Ann Arbor.
These documents will be presented in a new window.