Diffusion, Use and Impact of the Internet in Hong Kong: A Chain Process Model


  • Jonathan J. H. Zhu,

    Corresponding author
    1. Professor in the Department of English and Communication at City University of Hong Kong. He was formerly Associate Professor in the Department of Communication Sciences at University of Connecticut. His work has appeared in Journal of Communication, Human Communication Research, Public Opinion Quarterly, International Journal of Public Opinion Quarterly, Political Communication, Journalism & Mass Communication Quarterly, Journalism Monographs, and elsewhere.
    Search for more papers by this author
  • Zhou He

    Corresponding author
    1. Associate Professor in the Department of English and Communication at City University of Hong Kong. He was formerly Associate Professor in the School of Journalism and Mass Communication at San Jose State University. His work has appeared in Journalism and Mass Communication Quarterly, International Journal of Public Opinion, Journalism Monographs, Intercultural Communication Studies, Media, Culture and Society, and Gazette
    Search for more papers by this author

Address: Department of English and Communication, City University of Hong Kong, 83 Tat Chee Avenue, Kowloon, Hong Kong. Phone:(852)2788-7186 Fax: (852)2788-8894.

Address: Department of English and Communication, City University of Hong Kong, 83 Tat Chee Avenue, Kowloon, Hong Kong. Phone: (852)2788-7186 Fax: (852)2788-8894.


Hong Kong appears to be a dream venue for the Internet as a mass medium: There are a well-developed telecommunications infrastructure, a population with both financial resources and bilingual (Chinese and English) abilities, and a legal environment in which there is virtually no government regulation of content. However, recent experience with the slow adoption of other new media in Hong Kong, including cable TV and interactive TV, has sounded a cautionary note about the potential use and impact of the Internet in this technologically sophisticated city. Based on a telephone survey of 1,000 adult residents, this paper examines the adoption, use, and social impact of the Internet in Hong Kong using a chain process model that was initially developed by Dutton, Rogers, and Jun (1987) for research on home computing. The results show that Internet adoption is affected by a full range of factors, including one's personal characteristics, socioeconomic status, socio-cultural settings, and perceived compatibility of the Internet. On the other hand, Internet use is primarily affected by socioeconomic status and perceived compatibility. The study also found that both adoption and use of the Internet have observable impact on leisure activities and concerns for privacy and other Internet-related negative consequences. However, these effects are not overarching but rather confined to specific attitudes and behavior.


Hong Kong offers almost all the necessary and favorable conditions available for the new medium. Technologically, there is a well-developed telecommunication infrastructure in Hong Kong, with a virtually universal coverage of household telephone, more than 150 Internet Service Providers (ISPs), and half a dozen Broadband Service Providers (BSPs). Financially, the residents are among the wealthiest in the Asia-Pacific region. In addition, the majority of the population is bilingual (i.e., speaking Chinese and English). Finally, there is virtually no government regulation of Internet content.

However, the slow adoption of cable TV and interactive TV in Hong Kong suugestions that predictions of rapid diffusion, popular use, and penetration of the Internet in Hong Kong should be made with care. With a highly concentrated population served by only four broadcast channels, Hong Kong is supposed to be an ideal playground for cable TV. Nevertheless, after eight years of vigorous marketing, cable TV has penetrated only about a quarter of the households, which is among the lowest in all major cities in the Asia-Pacific region. Interactive TV is another case in point. In 1998, iTV, the first interactive TV service in the world, was launched in Hong Kong to provide news, video on demand, education, banking, sports, and other online services. As of today, only about 2% of the households have embraced this cutting-edge medium. Although there has been no systematic research on the slow diffusion of cable TV or iTV in Hong Kong, anecdotal evidence suggests that many viewers see no compelling incentives to adopt cable TV or iTV, as they feel that their needs are adequately served by the terrestrial TV, or that the new media are not a substantive replacement. It appears that, while technological and financial resources are necessary conditions, audience perceptions are a sufficient condition for the adoption and use of any new media technology.

The current study explores whether audience perceptions also play a part in the adoption and use of the Internet in Hong Kong. Although research on the Internet has proliferated in recent years, there is a lack of insightful theoretical framework to guide such research. We have, therefore, turned to a conceptual framework developed by Dutton, Rogers, and Jun (1987) for research on home computing, which is highly relevant and insightful for research on the Internet.

Theoretical Framework

The framework draws on a meta-analysis of 11 survey studies on the adoption, use, and impact of home computing among U.S. households. While it focuses on home computers, we think that the framework is largely applicable to research on the Internet. First, the notion of “home computing” involves use of personal computers at home for a wide range of activities including not only computing (e.g., word processing, family financial planning, and other office applications) but also communication and media functions (e.g., online communication and entertainment) that were precursors of today's Internet activities. Second, situated at a time when 18% of the U.S. households adopted home computing, the framework summarizes the theoretical reasoning of the early stages of the diffusion of home computing that are about where the Internet is in Hong Kong. Third, unlike most other studies on the home use of computers that take a single theoretical perspective, most noticeably diffusion of innovation (Rogers, 1983) or uses and gratifications (Rubin, 1994), the Dutton, Rogers, and Jun framework explicitly underscores the causal links among diffusion, use, and social impact of home computing and thus integrates these processes into a unified framework. Figure 1 below summarizes the exogenous, intervening, and dependent variables proposed by Dutton, Rogers, and Jun:

Figure 1.

A chain process model of the adoption, use, and social impacts of home computing.

Home computing involves a three-stage process: individual socioeconomic and demographic characteristics, perceptions and attitudes, socio-cultural setting, and hardware and software features serve as independent variables having a direct impact on (a) the adoption of home computers, which in turn determines (b) the use of home computing, which in turn affects (c) a wide range of perceptions and behavior including learning and education, family functioning, leisure activities, work from home, household routines, privacy, civil liberties, and property rights. The 11 survey-based investigations reviewed by Dutton, Rogers, and Jun (1987) have provided supporting evidence, in varying degrees, for some portions of the model.

Adoption of home computing. Of the three stages, adoption has received the most extensive research. The results consistently show that education is the single most significant individual contributing factor, followed by income, age and occupation. Also important is the socio-cultural setting, such as family structure (e.g., with children in the home), social networks (e.g., friends' owning a home computer), and use of computers at work. Only a few studies have examined, with mixed evidence, the impact of perceptions of technical features of home computing, such as difficulty of use and human compatibility. The observed effects of SES on the adoption of home computing bear immediate policy implications for the knowledge gap, inequality of access, and other related social issues that differentiate adopters and nonadopters.

However, as Dutton, Rogers, and Jun (1987) note, we know too little about the basis for the impact of SES, which leads them to the formulation of three hypotheses for future research: Members of higher SES are more likely to adopt home computing because of (a) the interactive nature of computing that demands more active cognitive skills, (b) the information-based job nature and the education-oriented values of people with higher SES, or (c) the high price of personal computers. While the last factor might be transient as the price of personal computers is falling to a level affordable to families of lower SES, the first two could represent some enduring effects.

These findings and corresponding hypotheses appear to be readily applicable to the adoption of the Internet, which requires adoption of home computers as one of the prerequisites. However, the two adoption processes differ in a number of respects. For example, while the adoption of home PCs is largely a family decision, the adoption of the Internet is an individual behavior. Also, the adoption of home PCs is more financially demanding than the adoption of the Internet, which implies that socioeconomic status of individuals may play a less critical role in the adoption of the Internet.

Use of home computing. This is an under-studied area. Most of the studies reviewed by Dutton, Rogers, and Jun (1987) provide only descriptive statistics of time allocated to different types of computing in the home. Overall, the studies have found that home computing is primarily used as an “instrumental” tool (e.g., for work, word processing, education, home budgeting, etc.). Use of home computing for entertainment (e.g., video games) is much less frequent. A longitudinal study has found that use of home computing becomes a constant once the novelty of home computing wears off (Venkatesh & Vitalari, 1986).

Dutton, Rogers, and Jun (1987) conceptualize that computer use involves two dimensions: amount of time and diversity of use. They further offer two competing hypotheses about the trend in diversity of use. On the one hand, home computing may be used for an increasingly large variety of purposes as computing hardware and software become more sophisticated. On the other hand, the range of home computing activities may become more concentrated for each user as the novelty of the innovation fades. However, none of the studies reviewed has attempted to document the range of diversity, let alone test the long-term trend. Also absent is an investigation of factors that facilitate or prohibit the use of home computing. Is the use of home computing affected by the same variables that determine the adoption of home computing? One may expect to see a diminishing role of SES because most users (i.e., early adopters) are from the same sectors of high SES.

Research on the use of the Internet has largely followed the path of the early studies of home computing. Reports on Internet usage are widely available, mostly from online measurement companies. However, we are yet to find causal explanations for the observed variations in user behavior. Diversity of Internet surfing is an entirely new concept. The current study aims to explore factors that influence both the amount of online time and the diversity of usage.

Social impact of home computing. Some of the studies reviewed by Dutton, Rogers, and Jun have found a (limited) negative impact of adoption and use of home computing on family functioning, such as more family conflict (Rogers, et al., 1982), less time with family and friends (Vitalari, et al., 1985), and more time alone (Venkatesh & Vitalari, 1986; Vitalari et al., 1985). The evidence is not consistent, though, as others have found no impact on social interactions (Caron, et al., 1983).

Adoption and use of home computing has also been found to affect leisure activities negatively. Leisure activity is the most frequently studied area of social impact. For example, adopters watched less TV, devoted less time to sports or other outdoor recreation, and engaged less in social interaction (e.g., Danko & MacLachlan, 1983; Rogers et al., 1982; Venkatesh & Vitalari, 1986; Vitalari et al., 1985). However, most of the studies rely on adopters' self-reports of time spent in these leisure activities, instead of comparing leisure time allocation between adopters and nonadopters.

Learning and education is an area in which most of the studies reviewed have found positive contributions made by the adoption and use of home computing. However, the findings are also mostly based on adopters' perceptions that home computing has been helpful to learning. Finally, there is inadequate research on three other areas of social impact: Work at home, household routines, and privacy, civil liberties and property issues.

In summary, the studies reviewed by Dutton, Rogers, and Jun (1987) are both incomplete and inconclusive. Several key variables and relationships identified, such as diversity of use, and the relationship between adoption and use, have received little empirical investigation. While the chain process model necessarily requires longitudinal studies, all but one of the studies involves a one-time snapshot, and thus the observed relationships might be temporal (i.e., the found impact might evaporate as the process evolves). For example, Dutton, Rogers, and Jun have noted that the strong relationship between education and adoption of home computing might be fading away as the cost of home computers keeps declining. Finally, although the model posits a multivariate process in which a host of individual, family, and social factors affect individually or jointly the diffusion and use of home computing, most of the studies employ a bivariate analysis that compares adopters and nonadopters. Consequently, some of the differences between adopters and nonadopters of home computing might have been caused by individuals' demographic characteristics or socioeconomic status. Despite the realization of the need for rigorous and comprehensive research designs and conceptualization, many studies of the Internet conducted recently have largely inherited the conceptual and methodological limitations of previous research on technological innovations.

The current study aims to provide a complete test of the chain process model proposed by Dutton, Rogers, and Jun (1987). It involves a full set of the key variables in the model, including adoption status, adoption history, amount of use, diversity of use, impact on family functioning, leisure activities, and privacy and civil liberties, and the causal relationships among them. Multivariate analysis is employed to control for spurious and confounding effects. The data involved are from the first annual survey of a three-year panel sample. While still a snapshot at this point, the findings of the current study will be validated by the data from the subsequent waves of the longitudinal investigation.



Telephone interviews of adult residents in Hong Kong were conducted through a local university survey institute during the last week of December 2000. In the survey, interviewers placed calls, up to five times, to residential households whose telephone numbers were generated from a computerized random digit dialing procedure. An adult member whose age was between 18 and 74 was selected from each contacted household, based on the last birthday method, for interview. In doing so, we have purposely excluded from the survey teenage and younger children on the ground that the use and impact of the Internet among them involves a quite different set of questions and thus requires a separate study.

We have also used another selection criterion (i.e., Chinese-language speaking) to screen out disqualified respondents because one of our research questions requires us to examine the use of the English-dominant Internet among non-English speakers. Because Chinese speakers account for more than 95% of the population in Hong Kong, the generalizability of our sample is not compromised.

Our final sample consists of 1,007 adult Chinese, which translates into a sampling error of ±3% at the 95% confidence level. As calculated through the RR3 formula of the American Association for Public Opinion Research (2000)1, the response rate of the survey is 38%. Although somewhat low, the response rate is typical of telephone surveys in Hong Kong, a city of 6.8 million people who experience a dozen or more telephone surveys on a daily basis.

The sample has been weighted based on the joint distribution of age and sex in the population in December 2000 (Census and Statistics Department, 2001). Compared with the original samples, the weighted sample is more conservative (about 5–10%) in key statistics (e.g., the proportion of Internet users). All data reported below are based on the weighted sample.


Internet adoption. Following the Dutton, Rogers, and Jun framework, we have measured two aspects of Internet adoption: Adoption Status and Adoption History. Dutton, Rogers, and Jun define Adoption Status as comprising two categories (“Adopter” and “Nonadopter”). As Rogers (1995) has noted, adoption status is not a static state but evolves in the post-adoption process. We have, therefore, further divided “Adopter” into “Continuous Adopter” and “Discontinued Adopter,” and “Nonadopter” into “Potential Adopter” and “Continuous Nonadopter.” In particular, we have used two questions to construct Adoption Status. The first question asks if the individual is currently using the Internet (including web surfing, email, online chat, shopping, games, etc.). A person is classified as a Continuous Adopter if the answer is “yes”; as a Discontinued Adopter if the answer is “used it before, but not anymore,” and as a Nonadopter if a “no” answer is given. The second question asks a Nonadopter how likely the person is to use the Internet in the next year or so. A person falls into the Potential Adopter category if the answer is “very likely' or “somewhat likely,” and into the Discontinued Adopter category if the answer is “not likely” or “don't know.” For Adoption History, we have simply asked the Continuous Adopters to recall the year in which he/she started to use the Internet.

Internet use. Following Dutton, Rogers, and Jun's conceptualization of computer use that involves two dimensions (i.e., amount and diversity of time), we have asked the Continuous Adopters to estimate the minutes per week they spend on the following six Internet activities: (a) reading online news, (b) sending/receiving e-mail, (c) participating in online chat or discussion, (d) searching for work-related (or study-related for full-time students) information, (e) searching for personal interest information, and (f) playing online games or other entertainment. The resulting six measures constitute the first dimension of Internet Use (i.e., Amount of Online Time).

The six measures of online time also form the basis for the second dimension of Internet use (i.e., Diversity of Online Time), based on the equation for entropy in information theory (Shannon & Weaver, 1949):


where H is the summary score of diversity, Pi is the proportion of the total online time devoted to the ith online activity (i= 1 to 6 in the current study), log2 is the logarithm with 2 as the base, and the minus sign is to offset the negative values resulting from the logarithm transformation of percentages (i.e., less than 100%). The H-statistic has been effectively used in agenda-setting research (e.g., Chaffee & Wilson, 1977; Culbertson, 1992; McCombs & Zhu, 1995) to measure information diversity or issue diversity. As the equation shows, the H-statistic is not affected by the amount of online time but related to the distribution of the six online activities. The more evenly a user spreads his/her online time across the six activities, the higher his/her H score is. If a user uniformly allocates one-sixth of his online time to each of the six online activities, he/she will receive a maximum diversity score (-2.585), whereas if a user allocates all time to a single online activity, he/she will receive a minimum diversity score (0), regardless of how many minutes per week the user goes online.

Social impact. Of the six categories of social impact of home computing identified by Dutton, Rogers, and Jun (1987), three (Family Functioning, Leisure Activities, and Privacy, Civil Liberties, and Property Rights) have been measured in our survey. For Family Functioning, we have asked the respondents to estimate the number of times per week they (a) talk, (b) have dinner, (c) watch TV, (d) play, and (e) shop with family members. For Leisure Activities, we have asked the respondents to estimate the number of minutes per day they (a) watch TV, (b) listen to the radio, (c) read newspapers, (d) do exercises, and (e) talk with friends outside workplace (or school for full-time students). For Concerns for Privacy and Civil Liberties, we have asked the respondents (both users and nonusers of the Internet) to comment on the extent of negative aspects of the Internet, including (a) violation of privacy, (b) pornographic content, (c) violent content, (d) junk information, (e) undesirable people, and (f) Internet addiction.

Independent variables. We have asked about age, sex, marital status, and family structure (i.e., living with children) of the respondents to measure Personal Characteristics; educational level, occupation, and family income to measure their Socioeconomic Status; number of Internet users at home to represent Socio-cultural Setting; and questions about (a) how useful the Internet is for their work or study, (b) how compatible the Internet is with their lifestyle, (c) how easy it is to use the Internet, (d) how easy it is to describe the benefits of the Internet to others, and (e) how much the Internet enhances their social status to measure Technical Features and Human Compatibility. The last five questions were in fact developed from an inventory of Perceived Characteristics of Innovation (PCI) (Moor & Benbasat, 1991).

Descriptive Summary of the Sample

Diffusion of the Internet. Of the 1,007 families in our sample, 69% are “PC Homes” with at least one computer, 52%“Internet Homes” whose PCs are connected to the Internet, and 21%“Broadband Homes” whose Internet connection is via broadband. Using individuals as the unit of analysis, 40% of the 1,007 respondents are “Continued Users” of the Internet, 7%“Discontinued Users” who have stopped using the Internet (but 5% are likely to use it again), 12% are “Potential Users” who are “Very Likely” or “Somewhat Likely” to do so in the next 12 months, and 43%“Unlikely-Users” who will not use the Internet any time soon (Figure 2).

Figure 2.

Status of Internet adoption in Hong Kong, December 2000.

Internet users. These respondents have on average used the Internet for 3.4 years; two-thirds (65%) of these have used it for two years or more. They spend on average 629 minutes (or 10.4 hours) per week on the Internet, including 350 minutes at home, 262 minutes at their workplace or at school, and 17 minutes elsewhere. In comparison, the users spend 14.4 hours per week watching television, 7.5 hours listening to the radio, 6.2 hours reading newspapers, 2.6 hours doing physical exercise, 8.4 hours socializing with family members, and 8.2 hours socializing with relatives, friends or coworkers/classmates. The users spend the largest amount of time on searching for work- or study-related information (200 minutes per week on average), followed by receiving/sending E-mail messages (175 minutes), searching for personal interest information (104 minutes), and reading online news (90 minutes). The activities that attract the least time include online chat or discussions (52 minutes) and online games or other entertainment (35 minutes). The users allocate about half (48%) of their online time to local websites using the Chinese language, another 15% of the online time to overseas sites using the Chinese language, and the remaining 35% to non-Chinese language sites, local or overseas.

E-mail, online discussions, and online shopping. E-mail appears to be very popular among Hong Kong users, with 57% of them using it at least once per day and another 32% on a less frequent basis. Only 11% of the users do not use E-mail. On average, the users communicate regularly with 7 people, including friends, colleges, or business contacts in Hong Kong (46%) or overseas (21%), and family members or relatives living overseas (24%) or locally (10%). Participation in online chat or discussions is much less popular, with only 31% of the users having ever done so. Online chat and discussions on personal hobbies ranks at the top among online activities (by 12% of the users), followed by personal relations (12%), investment (8%), and government and politics (5%). The users prefer locally run chat rooms or BBSs to overseas sites by a 6:4 ratio. Only 26% of the users (or 11% of the sample) have ever bought something online, with an average expenditure of HK$ 4,150 (or US$ 532), including books, art crafts, food, movie tickets, CDs, electronic appliances, and computer hardware and software.

Internet skills. Of eight online tasks, the users can perform 4.8 on average. The most popular tasks include online searching (94%), followed by software/music download (83%), saving web pages (73%), online discussions (70%), and attaching files to E-mail messages (69%). Only one-third or fewer of the users can design personal web pages (33%), use the Internet phone (33%), or set up proxy servers (29%).

Non-users. As compared with the users, the non-users share a similar sex ratio, family income, and marital status, but are significantly older, less educated, and less likely to hold a professional or managerial job. In addition, non-users are more worried about the negative impact of the Internet. Among the non-users, the potential users are younger and more positive about the Internet than the unlikely users. Of various reasons cited for not using the Internet, lack of knowledge ranks at the top (by 34%), followed by lack of interest or need (25%), lack of computers (16%), lack of time (12%), and fear of the technology (9%).

Perceptional and behavioral differences. After controlling for age, sex, education, occupation, family income, and marital status, users and non-users demonstrate a number of significant differences in perceptions and behavior. Compared with non-users, users are more positive about the innovativeness of the Internet, less concerned about the negative impact of the Internet, more suspicious of the political empowerment of the Internet, view television and newspapers as less credible, and spend less time (about 20 minutes per day) watching television. On the other hand, users and non-users do not differ in their views about the need for regulation of the Internet, in perceived credibility of the Internet, major corporations, government, schools, science, and the church, and in time spent on newspaper reading, exercising, socializing, and sleeping.

Main Findings

Internet Adoption

Impact on adoption status. As described earlier, Internet Adoption Status is a nominal-scale variable with four categories: “Continuous Adopters,”“Discontinued Adopters,”“Potential Adopters,” and “Continuous Nonadopters.” We have, therefore, used multinomial logistic regression (MLR) for the analysis. Table 1 reports the results of the MLR analysis. The first column of the table shows the overall impact of each independent variable, as tested by the difference in X2 between a full model that includes all 13 independent variables and a “one-less” model that excludes a particular independent variable under test. As shown in the column, eight of the 13 independent variables, from all four blocks of hypothesized influences, are significant. Most noticeably are Number of Internet Users in Home (representing one's socio-cultural setting), Age, and Education. Also highly significant are three of the five PCI variables, namely Compatibility, Image, and Result Demonstrability of Internet Use.

Table 1.  Multinomial Logistic Regression (MLR) coefficients predicting group membership of Internet adoption status. Thumbnail image of

The last three columns of Table 1 are unstandardized MLR coefficients, each representing the impact of an independent variable on the probability of being in one group versus the probability of being in the baseline group (i.e., Continuous Adopter). For simplicity, we will use the more familiar OLS regression language to report and interpret these logit-based coefficients. For example, when we say “X has a significant impact on Y,” it means that X has a significant impact on Y to be in group A as opposed to being in group B. Also noteworthy is that a negative coefficient means a lower probability for an individual to be in a comparison group (i.e., a higher probability to become an Adopter). Because there are six pair-wide comparisons among the four groups, an MLR only contrasts 3 (= 4 - 1) groups, as shown in Table 1. To help to see the full picture, we have constructed, based on the information in Table 1, a new table showing all six pair-wide comparisons involving eight of the 13 independent variables that have a significant impact on Internet Adoption Status.

Table 2 reinforces the finding that Education, Age, and Number of Internet Users at Home are strong predictors of Adoption Status. The three variables are significant in every pair-wide contrast among the four adoption status groups. It is also important to note that the group-wide difference is always the largest between the two extremes (i.e., Continuous Adopters vs. Continuous Non-adopters) and smaller between adjacent groups, suggesting a linear order among the four groups. Of the three significant PCI variables, Compatibility is a generally strong predictor as it is significant in four of the six contrasts. The impact of Image and Result Demonstrability appears to be quite limited. Interestingly, Continuous Adopters are least likely to believe that Internet use will enhance one's social image.

Table 2.  Multinomial Logistic Regression (MLR) coefficients for group-wide contrast. Thumbnail image of

Impact on adoption history. While the above analysis of Internet Adoption Status involves the entire sample (N= 1,007), Adoption History is necessarily confined to Continuous Adopters (N= 401). Because our measure of Adoption History is a single question of number of years since adoption, we have used Ordinary Least Square (OLS) regression. As shown in Table 3, while an individual's socioeconomic status and socio-cultural settings strongly affect how early the person adopts the Internet, his/her personal characteristics or perceived characteristics of the Internet do not matter much.

Table 3.  OLS regression coefficients predicting Internet adoption history. Thumbnail image of

Specifically, better-educated, wealthier, professional-managerial job holders (including civil servants) and retired-unemployed (as opposed to students, blue-collar workers, self-employed and other occupation workers), and those with family members using the Internet are more likely to be earlier adopters of the Internet in Hong Kong. Of the five perceived characteristics of the Internet, only Image is significant, but in an opposite direction to what previous research on diffusion of innovation has suggested. That is, those who think Internet use can enhance their social image are more likely to adopt the Internet at a later stage.

Internet Use

Impact on amount of online time. As previously described, we have measured the number of minutes per week Hong Kong Internet users spend on six online activities. Theoretically, the six measures of online time are inherently interrelated as they are taken out of the zero-sum game of daily life (i.e., 24 hours a day for everyone). Empirically, the six variables in our data are weakly correlated, in either positive or negative directions, which does not warrant a summary index for the six time variables. Therefore, we have chosen to use the General Linear Model (GLM) regression that takes the six time measures as a collective set of dependent variables and regresses them simultaneously on all independent variables in the model. The interpretation of GLM regression results (Table 4) is essentially the same as in the case of OLS regression.

Table 4.  General Linear Model (GLM) coefficients predicting online time. Thumbnail image of

As the first row of Table 4 shows, Internet Adoption History does not have a significant impact on the amount of online time, except for the time on search for work/study-related information. In particular, a user with each additional year of adoption spends 43 minutes per week more on work-related information search. Also noteworthy is the negative impact of adoption history on online entertainment, which is marginally significant (p= .07). The evidence seems to suggest that earlier adopters are more likely to use the Internet for work and less likely to use it for fun.

Of various measures of socioeconomic status, Education has a significant impact on the six variables as a whole, which is mostly attributed to the impact on e-mail. Occupation is also a significant predictor of online time. Table 4 shows that Students and Retired/Unemployed respondents spend significantly more time on online chat or discussions than members of the baseline group (i.e., Self-employed and Others). Though not shown in the table, Students and the Retired/Unemployed also use online chat or online entertainment significantly more than Blue-collar Workers and Professional/Managerial Job Holders, respectively. Professionals and Managers also spend significantly less time than Retired/Unemployed reading online news, but significantly more time than Blue-collar Workers to search personal interest information.

In general, personal attributes such as Age, Sex, Marital Status, and Family Structure do not have any significant impact on the amount of online time. Two exceptional cases are the impact of gender on search for personal-interest information and online entertainment, where male users spend significantly more time. Finally, one's socio-cultural settings, as measured by the number of Internet users at home, does not have any impact on the amount of online time.

Of the five PCI variables, three appear to have some significant impact on the amount of online time. Those who perceive the Internet to be compatible with their work/life style spend more time reading online news or searching for work/study-related information. Those who think it easy to explain the merits of the Internet to others devote more time to online chat or discussions. Unexpectedly, those who perceive the Internet to be easy to use actually spend less time online, especially on searching for personal-interest information.

Overall, the model provides an inadequate fit to the data, with only about 5–6% of the variance in the amount of online time accounted for.

Impact on diversity of online time. As shown in Table 5, Diversity of Online Time is significantly affected by Occupation, Age, and Perceived Compatibility and Result Demonstrability. In particular, Students, the Retired/Unemployed, and Professionals/Managers, younger users, those who think the Internet to be compatible with their work/life study, and those who feel it easier to explain the benefits of the Internet to others, are more likely to allocate their online time evenly across different applications. These four variables together explain almost one-fourth (22%) of the variance in Diversity of Online Time, a much better fit than in predicting Amount of Online Time.

Table 5.  OLS regression coefficients predicting diversity of online time. Thumbnail image of

Social Impact of Internet Use

The common practice in testing social impact of the Internet is to compare users and nonusers, with or without controlling for individual characteristics. The approach fails to take the impact of adoption history into account. To test both adoption and use of the Internet simultaneously, we have created a new independent variable (called “Adoption and Use”) with five categories: (a) Nonusers (N= 606), (b) Later Adopters/Light Users (N= 121), (c) Early Adopters/Light Users (N= 87), (d) Later Adopters/Heavy Users (N= 78), and (e) Early Adopters/Heavy Users (N= 108)2. The variable is entered as the main between-subjects factor (i.e., a nominal scale independent variable) into three GML regressions, each involving Family Functioning, Leisure Activities, and Concerns for Privacy and Civil Liberties as the dependent variable, respectively, while controlling for socioeconomic status, individual characteristics, socio-cultural settings, and perceived characteristics of the Internet.

Impact on family functioning. As Table 6 shows, Adoption and Use appears to have little impact on Family Functioning that involves dining, chatting, playing, watching TV, or shopping together with family members. Of 20 contrasts between users and nonusers, only one reaches statistical significance at the .05 level, in which the Later Adopters/Light Users play with family members more frequently than the Nonusers. Within the four user groups, the Later Adopters/Light Users spend more time playing with family members, but less time dining with family member, than the Early Adopters/Light Users (not shown in Table 6). No other difference has been found in any other contrasts among the four user groups.

Table 6.  General Linear Model (GLM) coefficients predicting family functioning. Thumbnail image of

Of the 13 covariates, Marriage Status is the most significant determinant of Family Functioning. As can be expected, those who are married spend significantly more time with family in daily activities. Socioeconomic Status, Socio-cultural Settings, and Perceived Characteristics of the Internet have, in general, little impact on Family Functioning.

Impact on leisure activities. Adoption and Use also has some impact on two of the five leisure activities analyzed (Table 7). First, members of three user groups, namely Early Adopters/Light Users, Early Adopters/Heavy Users, and Later Adopters/Light Users, spend significantly less time (about 45–50 minutes per day) watching TV than Nonusers, other things being equal. Second, both Later Adopters/Light and Later Adopters/Heavy Users exercise significantly less (about 30 minutes per day) than Nonusers. No difference has been detected between any pair of the groups in three other leisure activities such as listening to the radio, reading newspapers, and socializing with friends. Overall, frequency of Internet use appears to carry a little more weight than history of Internet adoption.

Table 7.  General Linear Model (GLM) coefficients predicting leisure activities. Thumbnail image of

Of the covariates, Marital Status again demonstrates the most significant impact, with those married watching less TV but spending more time on all other leisure activities. Interestingly, three of the five Perceived Characteristics of the Internet are related to Leisure Activities. However, the impact is not in a uniformed direction.

Impact on concerns for privacy and civil liberties. Adoption and Use has a significant overall impact on the six measures of Concerns for Privacy and Civil Liberties (Table 8). In general, users are less concerned about disclosure of privacy, exposure to violent or pornographic content, making bad friends, or becoming addicted from using the Internet. Specifically, Later Adopters/Light Users are significantly less concerned about bad friends, pornography, violence, or addiction, Later Adopters/Heavy Users are less concerned about privacy, pornography, or violence, and Early Adopters/Heavy Users are less concerned about bad friends or violence. The difference between Early Adopters/Light Users and the Nonusers is not significant in any of the five areas. The four user groups are, on average, more concerned about receiving junk information than the Nonusers, although none of the user groups significantly differs from the Nonusers on a pair-wise comparison.

Table 8.  General Linear Model (GLM) coefficients predicting concerns about privacy and civil liberties. Thumbnail image of

Of the covariates, Sex, Perceived Advantages of the Internet, and Perceived Image of Internet Use have a significant impact on Concerns for Privacy and Civil Liberties. Interestingly, women are less concerned about privacy, Internet addiction, or making bad friends online. On the other hand, those who think the Internet can enhance their work/study and life are more concerned about becoming addicted or getting junk information. Those who think the Internet can enhance their social image are more worried about invasion of privacy, and exposure to pornographic or violent content. Overall, however, the model has explained only 5% or less of the variance.

Conclusion and Discussions

The current study investigates the adoption, use, and attitudinal and behavioral consequences of the Internet among 1,007 adult residents in Hong Kong. The study has been guided by a theoretical framework developed by Dutton, Rogers, and Jun (1987) for research on home computing. While the framework identifies key variables and relationships among the variables involved in the adoption, use, and impact of home computing, much of the model has not been empirically tested, and still less has been extended to networked computing at work or home (i.e., the Internet). In the current study, we have provided a complete test of the full model with a number of interesting findings, which are summarized in Figure 3.

Figure 3.

Summary of empirical test of the chain process model of adoption, use, and social impact of the Internet.

As shown in Figure 3, we have found that all four blocks of independent variables identified by Dutton, Rogers, and Jun, namely socioeconomic status, personal attributes, socio-cultural settings, and perceived technical features, have strong effects on an individual's past, current, and possible future adoption of the Internet. Interestingly, there is always a single most dominant predictor in each of the blocks. For example, Education is the most dominant determinant of Adoption Status in the block of socioeconomic status variables. Likewise, Age is the most influential among the variables of personal characteristics, and Perceived Compatibility among the block of perceived characteristics of the Internet. Number of Internet Users at Home, however, as the only measure of socio-cultural settings, has emerged as the strongest predictor of all four blocks of independent variables. In short, 40% of the adult population in Hong Kong have adopted the Internet, who are well educated, young, live with other Internet users at home, and view the Internet as compatible with existing work/study routine or lifestyle.

Altogether, the findings fully support the Dutton, Rogers, and Jun model by not only underscoring the importance of socioeconomic status and personal characteristics in the adoption process, but also demonstrating the critical role of socio-cultural settings and perceptions of technical features and compatibility. As we contended at the outset of the study, socio-cultural settings and perceived compatibility might also be the key to explaining the slow diffusion of cable TV and the failure of interactive TV in Hong Kong. In short, socio-cultural settings and perceived compatibility may have proven to be the two most essential determinants of new media adoption in a society like Hong Kong where there is no lack of technological infrastructure or financial means.

While Adoption Status is simultaneously affected by all four blocks of independent variables, our second measure of adoption, Adoption History, is primarily related to two of the blocks: socioeconomic status (as represented by Education) and socio-cultural settings (by the Number of Internet Users at Home). Personal characteristics (e.g., Age) and perceived characteristics of the Internet (e.g., Compatibility of the Internet) do not appear to have any impact on the exact timing of an individual's adoption of the Internet. The absence of Age's influence on Adoption History is readily understandable because the impact of Age is cancelled off by two opposing forces: Young people are more likely to be adopters (e.g., 75% of those between 18 and 25 are adopters, as compared to only 14% of those 45 or older) on the one hand, but they were too young a few years back to be an early adopter (e.g., the average number of years in using the Internet is 3.2 years for the 18–25 cohort but 3.3 years for the 45+ cohort) on the other. This confirms Dutton, Rogers, and Jun's notion that some of the causal relationships are necessarily transient and thus should be understood along with the time dimension.

Our data also have shown that socioeconomic status and perceived compatibility of the Internet affect both operationalizations of Internet use: Amount of Online Time and Diversity of Online Time. In fact, the current study is the first attempt in media research literature to measure and analyze empirically the diversity of new media use. The results have demonstrated that our operational measure of diversity is quite robust and sensitive to causal influences. With all the careful treatments in measurement and analysis, however, we have not found, quite surprisingly, any evidence for the hypothesized link between adoption and use of the Internet. That is, early adopters spend just as much time as later adopters. The evidence is both clear-cut and consistent because neither of the two measures of Internet adoption (i.e., Adoption Status and Adoption History) has any significant impact on either of our measures of Internet use (i.e., Amount of Online Time and Diversity of Online Time). In addition, this finding is consistent with our study of the Internet in Mainland China that has also shown no causal connection between adoption and use (Zhu & He, 2001). That study can help to answer the question of why Internet adoption has no impact on Internet use because the study has involved several additional variables, such as Perceived Needs for Internet (PNI), to help uncover that adoption and use of the Internet are two distinct processes subject to different influences (Zhu & He, 2001). This seems to make sense because Internet adoption requires a one-time investment in the hardware, no matter how costly it may be, whereas Internet use demands time and attention on a continuous basis. The latter is a far more costly commitment. Given the fact that our models have explained only 5 to 15% of variance in the amount of online time and 22% of the variance in the diversity of online activities, there is clearly a need to explore other causes of Internet use. Webster and Wakshlag's theory of structural determinism of television viewing (1983) points to a promising direction: the availability of free time, which has proven to be the most important determinant of the amount of time viewers spend on television and may function the same way for use of the Internet.

To put our findings in perspective, we can divide the causal factors of Internet adoption and use into the following categories. Socioeconomic status appears to be a uniformly influential predictor that affects all four aspects of the adoption and use process. Socio-cultural settings are primarily a driving force for Internet adoption. Conversely, perceived human compatibility and other features of the Internet are primarily determinants of both the amount of Internet use and the diversity of Internet use. Finally, personal characteristics split their influences in part of the adoption process and part of the use process. This categorization perhaps provides some additional explanation of why adoption and use of the Internet are two independent processes.

Finally, we have tested the impact of the Internet on Hong Kong people's attitudes (toward Internet-related issues) and behaviors (in both leisure activities and family functions). Social impact of the Internet has always been the center of the debate within and outside the media research community. Our study has shown that adoption and use of the Internet do have some discernible impact on users. However, the impact is not overarching but rather confined to certain attitudinal and behavioral responses. For example, we have found that Concerns for Privacy and Civil Liberties are affected by both adoption and use of the Internet whereas Leisure Activities are largely influenced by Internet use but not by Internet adoption. Family Functioning is, however, largely unrelated to either adoption or use of the Internet. In interpreting these findings, we need to keep at least two points in mind. First, because we have included in the test a wide range of demographic, socioeconomic, family, and perceptional covariates, it is understandable to find the Internet to have weaker effects than what has been found in other studies that simply compare users and nonusers without controlling for the confounding impact of other variables. As the near century-long history of media effects research has shown, “strong effects” based on bivariate analysis are almost always destined to evaporate after rigorous controls are exercised. Second, given the fact that less than half of the population in Hong Kong has embraced the Internet, it may be too early to detect the changes brought up by the new media technology in people's deeply rooted views and daily routines.


An earlier version of this paper was presented at the Annual Conference of the International Communication Association, Washington, DC, May 2001. The study was supported by a Competitive Earmarked Research Grant (CityU 1152/00H/9040555) from the University Grants Committee of Hong Kong.


  • 1

    The formula defines response rate as the number of completed interviews divided by the number of interviews, non-interviews (refusal and break-off plus non-contacts plus others), and the estimated eligible cases among status unknown cases.

  • 2

    We use Adoption History = 3 (i.e., year of 1998) and Total Amount of Online Time = 600 (i.e., 10 hours per week) as the cutting points to create the four categories of users. Those who started to use the Internet in 1998 or later are considered “Later Adopters,” and those who use the Internet 599 minutes per week or less “Light Users.”