| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
RESEARCH ARTICLE |
1 Department of Sociology and Center for the Study of Aging, University of Vermont, Burlington.
2 University Honors College
3 Department of Human Development and Family Sciences, Oregon State University, Corvallis.
Address correspondence to Dr. Stephen J. Cutler, Department of Sociology, University of Vermont, Burlington, VT 05405. E-mail: Stephen.Cutler{at}uvm.edu
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
Methods.Data are drawn from the September 2001 Current Population Survey and its supplement on computer and Internet use (
). Patterns of age differences in home computer use are examined using Multiple Classification Analysis with controls for several measures of compositional variability (employment status, marital status, Hispanic origin and race, gender, family income, living arrangements, education, and number of disabilities).
Results.Even though home computer availability declines steadily with age, a portion of this zero-order relationship is due to composition effects, with their greatest impact occurring at the oldest ages. Among persons living in households where a computer is available, use also declines with age, but these age differences in computer use are less due to compositional differences.
Discussion.The lower rates of home computer availability and use exhibited by older persons may be accounted for to some extent by compositional characteristics negatively associated with availability and, to a lesser extent, use of computers, although unmeasured attitudes, experience, and support undoubtedly play a role.
The advent and application of personal computer technology and Internet communications has been sweeping. According to A Nation Online (U.S. Department of Commerce, 2002
, pp. 1–2), 54% of the U.S. population is online and at least 45% is using e-mail. Although computer usage is increasing, regardless of sociodemographic, racial, or income characteristics, analyses of bivariate effects suggest that all forms of utilization continue to be strongly but inversely associated with age (U.S. Department of Commerce, 2002, p. 14). Peak utilization occurs among people in their late teen years, declines slightly among those in their early 20s, and then remains fairly stable until people reach their mid-50s, with approximately 70% of the populace availing themselves of computers for one purpose or another. After that age, computer usage appears to decline again and continues downward through age 80, the oldest age at which published information is available.
Perhaps because of the recency of computer technology, or as a consequence of the many small-scale descriptive investigations of the linkage between age and computer use, older persons are sometimes portrayed as being on the low or nonuse side of a "digital divide." A more considered perspective is held by a few investigators who have conducted more rigorous research (Czaja, 1997; Czaja, Guerrier, Nair, & Landauer, 1993; Fox, 2001). For instance, Czaja (1997) pointed out that the data often dispute the stereotypes about anxiety and attitudes irrevocably interfering with use. Along with a call for additional research, she noted the incontestable relevance of computer technology for accessing information regarding health care, financial management, communications, smart houses, social support, recreation, education, personal security, cognitive maintenance, and rehabilitation (Czaja, 1997, 2002). These applications may be particularly beneficial to older adults as they move further into old age and become increasingly homebound.
As we note in the paragraphs that follow, a major thrust of research has been to elucidate the roles of attitudes, training, and experience on the use by older persons of technology in general and computer use in particular. Other studies have focused on the descriptive documentation of age differences and other sources of variation in the availability or use of computers. In the present study, we add to our understanding of the reasons for age differentials in access to and use of computers by considering whether the compositional characteristics of older age cohorts account for their lower rates of use. Within a multivariate framework and by using a recent, large-scale, nationally representative data set, we examine age differences after we control for a variety of compositional factors (e.g., lower levels of education and income) that are known to be associated both with age and with access to and use of computers. By looking at the effects of compositional variables, we should also be in a position to comment on the implications of aging and cohort flow for the probable directions of prospective changes in computer utilization. That is, will there be a generational shift in use of computer technology and will existing patterns be time limited? If we are to understand the likely nature of changes over time, we must make every effort to sort out interactions between age and its associated compositional characteristics.
Background
How attitudes about technology affect willingness to use computers has been the subject of much research. It is frequently asserted that older persons have less favorable attitudes and greater anxiety about computers and information technology than younger persons and are therefore less apt to make use of technology when it is available (Ryan, Szechtman, & Bodkin, 1992). It is assumed that older people feel more uncomfortable and less competent and are therefore more resistant to using newer technologies than their younger counterparts (Czaja & Sharit, 1998). In light of a number of constraining factors—especially the relatively small sample sizes available for assessing attitudes—it is not surprising that the role of attitudinal factors is not well understood (Czaja & Sharit, 1998). In a very early article on attitudes toward computers, Kraus and Hoyer (1984) suggested that age, gender, and experience affect attitudinal dispositions toward computer technology. Czaja and Sharit (1998) examined attitudinal factors among 384 adult users, aged 20 to 75, and reported age effects in such areas as sense of comfort, efficacy, dehumanization, and control. They also noted that both gender and the nature of computer tasks made a difference insofar as attitudinal outcomes were concerned. Their principal conclusions were that negative attitudes not withstanding, feelings toward computers are modifiable, and understanding how age interacts with attitudes and usage is a first step in designing intervention strategies. Then, too, Czaja (1997) and colleagues (Czaja et al.,1993) noted that perceived usefulness and relevance affect the receptiveness of older users. A number of other investigators (Charness, Kelley, Bosman, & Mottram, 2001; Dyck & Smither, 1994; Jay & Willis, 1992; Mackie & Wylie, 1988) maintain that existing attitudes, anxiety levels, and competencies are subject to change and can be positively modified. For example, in a study of 100 community-dwelling people between the ages of 57 and 87, Jay and Willis (1992) reported that experience and training exert salutatory effects on negative attitudes and that older persons who have access to opportunities to hone their skills also demonstrate more positive attitudes.
In addition to whatever reticence older persons themselves might hold, another factor likely to affect computer usage is perception by others. Among a convenience sample of 200 undergraduates and science museum visitors, Ryan and colleagues (1992) assayed perceptions of the probability of success for computer training of a hypothetical older computer user. Age was perceived as an inhibitory factor, and younger persons felt that older persons who were using computers were atypical of their age group. The impact of negative perceptions of older persons' adaptability to and training for computer technology was also made clear by Edwards and Engelhardt's (1989) finding that the more supportive the exposure to computer training, the greater the comfort level of older users. Thus, training methods are linked to the acquisition of skills (Czaja, 1997), although, to our knowledge, to date no age-specific training models successfully muting age differences have been reported.
Interestingly, people with disabilities may have the most to gain from computer technologies yet are among the least likely to utilize them. In a descriptive analysis using the computer and Internet use supplement to the March 1998 Current Population Survey, Kaye (2000) found that among older Americans with work disabilities, approximately 10% owned computers compared with 25% of those without such disabilities. Among all persons with disabilities, elderly people, persons with low incomes or modest educations, and African Americans and Hispanics were less likely to utilize computer technology. In terms of gender differences, men and women with disabilities were equally as likely to have a computer in the household and to use it to access the Internet. Kaye's research dealt only with work disabilities, but a recent report from the U.S. Department of Commerce (2002) revealed that older persons (aged 60+) with other disabilities do indeed have access to home computers at rates below the overall national average. Whereas 56.5% of American households had a computer in 2001, only 30% of older persons who are blind or suffer severe vision impairments, 35% who are deaf or have severe hearing impairments, 29% of those with difficulty walking, 26% of those who find it hard to leave home, and 25% who experience difficulty typing had a computer at home (U.S. Department of Commerce, 2002, Table 7-5).
Descriptive analyses of Internet access also disclose gradients by age. Data from a 2000 survey of more than 26,000 adults show that persons over the age of 65 make up only 4% of Internet users, despite comprising 13% of the U.S. population, and they are overrepresented among the more adamant Internet holdouts (Fox, 2001). Fox concluded that there is a "gray gap," that is, older persons are not following the lead of their younger counterparts, not using computers, and not feeling as though they are missing out (Fox, 2001, p. 2). Among that narrow slice of the older population with connectivity—accounting for approximately 15% of those over the age of 65—income and education, plus cajoling by other family members, appear to be causal factors in their Internet usage. Approximately 60% of those older persons using the Internet are men, 74% are married, and over three quarters (76%) are college educated. Approximately 25% of all older users have household incomes over $75,000 annually, compared with only approximately 8% of the overall older population reporting incomes in that range.
From the available data, then, it is evident that older persons are less likely than their younger counterparts to live in households where a computer is present, and they are less likely to use a computer when it is available (U.S. Bureau of the Census, 2001; U.S. Department of Commerce, 2002). It is also clear that many of the compositional characteristics of the older population are associated with lower rates of access and use. For example, persons with incomes over $75,000 are three times more likely to have a computer in their home than persons with incomes below $25,000. Persons with a bachelor's degree or more are four times as likely to have a computer as persons with less than a high school diploma. Being married, being in the labor force, and living in multiple-person households are all related to access to a home computer (U.S. Bureau of the Census, 2001). However, the extent to which age differentials in access to and use of home computers can be accounted for by such compositional characteristics remains an unanswered question. To our knowledge, only one study (Morrell, Mayhorn, & Bennett, 2000) has modeled the effects of age on computer use net of the effects of compositional factors. On the basis of a sample of 550 persons aged 40 or older in southeastern Michigan, Morrell and colleagues found that the effect of age on World Wide Web use was diminished, but still significant, after they controlled for access, demographic, and socioeconomic factors.
The purpose of this article, therefore, is to shed additional light on questions of computer availability and computer use in old age and to bring findings from a rather substantial data set to bear on the issues. We contend that it is important to move beyond the descriptive zero-order analyses predominant in the literature in order to examine a number of potential influences simultaneously and to try to disentangle the effects of age from compositional factors characterizing diverse age cohorts. Specifically, our intent is to utilize a large, nationally representative data set—the September 2001 Current Population Survey (CPS) and its supplement on home computers and Internet use—to examine patterns of home computer availability and use. Our primary research question is, to what extent do compositional differences account for the global relationship between age and availability and use of computers? Though our data set is cross-sectional, meaning we can only examine age differences, the robust nature of the CPS sample is such that even a cross-sectional analysis is informative as it is based on by far the largest, most recent, and most representative sampling of computer usage.
|
METHODS |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
, for additional information about the CPS).
Sample Size, Missing Data, and Representativeness
The September 2001 CPS includes data on a total of 143,300 persons residing in 56,366 households. Because our interest is in home computer availability and use among adults, we restrict the analysis to persons who are 25 years of age and older. (This is also consistent with the Census convention that reports level of educational attainment—one of our principal control variables—for the population 25 and older, or approximately the age when formal education has been completed for much of the population.) Limiting the analysis to individuals 25 and older drops the sample size to 93,386 persons in 52,268 households. When missing data on all variables are taken into account, the final sample for the analysis is further reduced to 71,182 individuals in 40,264 households.
Because the exclusion of respondents with missing data reduces the size of the analysis sample to 76% of the full sample of persons 25 years of age and older, we present in Table 1 the results of descriptive analyses designed to assess whether any substantial biases are introduced by the elimination of persons with missing data. As is evident, the truncated analysis sample corresponds quite closely to the parent sample from which it is drawn. The gender distributions are identical, and the average of the absolute differences on race and education is less than.3%, with none of the individual differences exceeding.7%. Thus, these distributional comparisons show only negligible discrepancies between the samples and indicate that bias stemming from missing data is not likely to be problematic.
|
). Second, for each person in the household over the age of 3 years, the respondent was asked to indicate whether that person uses a computer at home (coded ). In order to allow for a detailed examination of age differences in home computer availability and use, we categorize age into fourteen 5-year groups, beginning with 25–29 and ending with 90 years of age and older.
Given our interest in determining whether zero-order age differences in home computer availability and use might be at least partially attributable to differences between age groups in terms of compositional characteristics, several control variables are used. These include factors identified in previous research such as gender, level of educational attainment (6 categories ranging from 0–8 years to graduate degree), family income over the past 12 months (14 categories ranging from less than $5,000 to $75,000 and over), and number of persons in the household, with an upper limit of five and over. Employment status is coded as a four-category variable (employed, unemployed, retired, or not in the labor force for other reasons), marital status is divided into six groups (married, spouse present; married, spouse absent; widowed; divorced; separated; or never married), and Hispanic origin and race includes five groups (Hispanic origin; White; Black; American Indian, Aleut, or Eskimo; or Asian or Pacific Islander). Finally, we have created a summary, additive measure of disabilities. Scores range from 0 (no disabilities) to 5 (five disabilities), based on the presence or absence of four physical conditions (blindness or a severe vision impairment even with glasses or contact lenses; deafness or a severe hearing impairment even with a hearing aid; a physical condition that substantially limits the person's ability to walk or climb stairs; or a condition that makes it difficult to type on an ordinary typewriter or traditional computer keyboard) and on whether the person has difficulty going outside the home alone (e.g., to shop or visit a doctor's office) because of a physical or mental health condition lasting 6 months or longer. We elected to use the additive index in the analysis rather than the individual disability measures after preliminary analyses showed that all measures were positively correlated with each other and all were negatively correlated with both home computer availability and use.
Analysis
Multiple Classification Analysis (MCA) was the principal data analysis method. This multivariate technique can be used to examine the relationship between a single predictor variable and a dependent variable (e.g., the relationship between age and home computer availability and use) or between each of a set of predictor variables and a dependent variable, simultaneously controlling the effects of the remaining predictors (e.g., the relationship between age and home computer availability and use after the effects of the several control variables are removed). No assumption of linearity is required. To determine the relationship between an independent and a dependent variable, MCA yields gross or unadjusted mean scores on the dependent variable for respondents in each category of the independent variable. With multiple predictors, MCA provides adjusted net scores, which are equivalent to the mean value of the dependent variable for each category of a predictor after the effects of the remaining predictors are controlled for. Eta and beta coefficients are available to assess the strength of relationships at the bivariate and multivariate levels, respectively, as are F tests to determine whether any given predictor (e.g., age) explains a significant proportion of the variance in the dependent variable before and after other predictors are held constant and to assess the significance of the combined effects of all variables.
|
RESULTS |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
|
Relationships between home computer availability, age, and our control variables are presented in Table 3. Among those who do have a computer at home, the relationships between personal use of the computer, age, and the control variables are shown in Table 4. In light of the 0 or 1 coding of the dependent variables, the unadjusted and adjusted mean scores are equivalent to the proportion of persons in each category who have a computer in their home and the proportion of persons in each category who use that computer.
|
|
As indicated earlier, our primary interest is to determine whether, and to what extent, lower rates of computer availability and use among the older population are due to compositional characteristics that serve to dampen the likelihood that a computer is at hand and is used by older persons. We have shown in Table 2 that age is associated with a variety of compositional characteristics (e.g., lower levels of education and family income and higher levels of widowhood) and in Table 3 that these characteristics are also associated with lower levels of computer availability. It is reasonable to assert that the effect of such compositional characteristics, then, is to depress rates of computer availability among the older population. At the same time, several of the control variables are, themselves, intercorrelated (e.g., widows are significantly more likely to reside in one-person households: 
, 
, and 
). Thus, we now seek to determine the patterns of age (and other) differences after we have controlled for the effects of the remaining variables.
It is clear from the column of data presenting adjusted mean scores in Table 3 that a portion, but not all, of the zero-order pattern of age differences in home computer availability is attributable to compositional effects, with the greatest impact being at the oldest ages. Among persons 90 years of age and older, for example, only 16.5% reside in households where a computer is available, but that figure could have been substantially higher—41.3%—if the characteristics of persons in the oldest age group had been comparable with persons in other age groups. Similarly, computer availability would be more than twice as high among persons aged 80–84, and 39% higher among persons aged 70–74. In general, the magnitude of the effect of adjusting for the control variables diminishes with successively younger age groups, so that by ages 55–59 there is only a 1% difference between the unadjusted and adjusted means. The overall magnitude of the reduction in age differences is also evident from a comparison of eta and beta; although still significant, the beta for the adjusted effects is.138 as compared with an eta of.317 for the unadjusted effects of age differences on home computer availability.
Even though all of the adjusted effects of the control variables remain significant, the magnitude of those relationships decreases and several of the major differences observed at the bivariate level are appreciably reduced. For example, after the remaining variables are controlled for, retirees are only marginally less likely than persons in other employment status categories to reside in households where a computer is available, and only a minor difference now separates widows from persons who are married with a spouse present. Ethnic and racial differences persist, albeit in diminished magnitude, and women are now slightly more likely than men to reside in households where a computer is available. Family income, household size, and education differences remain evident, although the betas are consistently smaller than the zero-order etas. With the effects of other variables controlled, only minor differences occur by number of disabilities. Overall, age proves to be the third strongest predictor of the availability of computers in the household, following family income and education, and, taken together, all of the variables explain 33.1% of the variance in availability.
The data in Table 4 present unadjusted and adjusted relationships between age, the control variables, and computer use for persons who reside in households where a computer is available. Taken at face value, the bivariate relationships show clear age differences in computer use. Advancing age is associated with lower levels of home computer use, with more precipitous declines occurring after ages 70–74. Retirees, persons who are widowed, and those of Hispanic origin are least likely to make use of computers in their homes, but women are slightly more likely than men to take advantage of computer availability. Use generally increases with family income and education but declines with increasing household size, which is a relationship running in the opposite direction of that seen for availability. Finally, the percentage of persons using a home computer declines as the number of disabilities rises.
An examination of age differences in computer use adjusted for the effects of the control variables shows that only a small portion of the original bivariate age effects is due to variation in compositional characteristics. This is particularly the case among the oldest age groups, whereas the effects of the control variables cease to dampen rates of use prior to ages 75–79. That the effects of the compositional characteristics play a limited role in explaining age differences in computer use is evident from the fact that the multivariate beta (.228) is actually higher than the bivariate eta (.215).
Comparisons of the etas and betas show that differences in computer use by employment status are reduced appreciably, as are differences by marital status and Hispanic origin or race, although to a lesser extent. If the effects of other variables are taken into account, computer use among retirees and widows is considerably higher than was the case when only unadjusted effects were looked at. Women continue to have higher rates of use than men, family income differences are greatly diminished, and the effect of household size—with lower rates of use associated with increasing household size—is strengthened. Level of educational attainment continues to have an effect on computer use, although it is clear that a substantial portion of the bivariate effects of disabilities can be explained by the other controls. Overall, the adjusted effects of each of the variables are statistically significant, and together they account for a significant proportion of the variance in computer use (R2 =.166). Education is the best predictor of computer use among persons living in households where a computer is available (
), followed closely by age (
).
|
DISCUSSION |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
The unadjusted mean scores for computer availability shown in Table 3 might lead one to conclude that there is a long and fairly level plateau from ages 30–34 to 50–54, followed by a decline that becomes even more pronounced beginning at ages 60–64. At first glance, then, these data appear to confirm what others have suggested. However, once key compositional factors such as gender, education, income, size of household, disabilities, employment and marital status, and race are controlled, a subtle but important shift occurs. A long, slow, linear decline does take place, but it is neither as rapid nor as severe as the bivariate analysis suggests. Rather, compositional factors do indeed appear to affect the pattern of age differences in the availability of home computers, and when they are partialled out, the drop-off at the older ages is not nearly as dramatic.
Among persons living in households where a computer is available, there are also clear and pronounced age differences at the zero-order level in the use of the home computer. For example, use declines from just under 90% among persons aged 25–29 to approximately 80% for persons aged 50–54, 63% for those aged 70–74, and on down to 39% among those aged 85–89. Of interest, though, is that the effect of controlling for compositional differences is not as consequential as it was in the analysis of availability of computers. To be sure, usage rates adjusted for compositional factors are higher, but this effect is limited largely to persons over the age of 85. In contrast to the findings for the analysis of computer availability, where compositional controls reduced the magnitude of the effect of age differences from an eta of.317 to a beta of.138, the adjusted beta for age differences in home computer use (.228) is slightly higher than the unadjusted eta (.215).
These results lead us to conclude that compositional effects do play an important role in explaining age differences in involvement with computer technology, but that their effects operate largely on whether a computer is available in the home rather than on the use of home computers. Older persons are less likely to have graduated from high school and college, and they are more likely to be at the lower income levels and to report a disability. They are more likely to be retired, widowed, and living alone. Because these and other characteristics are also associated with a reduced propensity to live in a household where a computer is present, it is not surprising that some portion of observed age differences in the availability of a home computer can be accounted for by compositional factors. That controlling for these same compositional variables has far less of an effect on age differences in computer use suggests the primacy of other, unmeasured factors in explaining variation in usage. We cannot directly test this hypothesis with the CPS data, but it may be the case that demographic and socioeconomic differences are of greater consequence in understanding the age patterning of access to home computers, whereas attitudes, experience, opportunities for training, and social support are more important in explaining usage patterns once a computer is available.
Another inference to be drawn from our results concerns usage patterns among older cohorts in the future. On the one hand, it might be argued that factors such as the persistence of income differences and higher rates of disability among future cohorts of older persons would work to dampen levels of access to computers at home. On the other hand, and leaving aside the use of computers in school or the workplace, our data show that over 70% of persons in their 30s and 40s live in a home where a computer is available and over 80% make use of the computer. The experience, prior training, and presumably the existence of favorable attitudes implied by these current levels bode well for use in the future. Moreover, our analyses show that education is the second best predictor of availability and the best predictor of use, and future cohorts of older persons will have appreciably higher levels of educational attainment than current cohorts (U.S. Bureau of the Census, 2000, Table 1a). Thus, as younger cohorts begin to age, it is quite possible that their computer usage will remain strong, and the patterns found among older persons in the future will resemble patterns found today among middle-aged adults.
Then, too, if it is correct that usage is related to relevance, the spread of technology into events and issues affecting older persons is apt to facilitate rather than impede usage. The advent of "smart environmental controls," computer communication, virtual communities, and supportive services is likely to have great appeal to older users and may serve to propel even greater computer use (Bass, McClendon, Brennan, & McCarthy, 1998; Cutler & Hendricks, 2001; Czaja, 2002; Czaja & Rubert, 2002). Technology promises new types of connectivity between older users and the larger world, including Web cameras to facilitate long-distance interaction, the World Wide Web as a source of information, and emerging applications of e-commerce (Cutler & Hendricks, 2001). The potential is immense and not likely to be anchored by any patterns presumed to exist among those who are currently old.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
Received for publication July 3, 2002. Accepted for publication January 8, 2003.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  S. A. Iverson, K. B. Howard, and B. K. Penney Impact of Internet Use on Health-Related Behaviors and the Patient-Physician Relationship: A Survey-Based Study and Review J Am Osteopath Assoc, December 1, 2008; 108(12): 699 - 711. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 K. Wild, L. Boise, J. Lundell, and A. Foucek Unobtrusive In-Home Monitoring of Cognitive and Physical Health: Reactions and Perceptions of Older Adults. Journal of Applied Gerontology, January 1, 2008; 27(2): 181 - 200. [Abstract] [PDF]
|
|
|
|
|
|
K. A. Smyth, J. H. Rose, M. J. McClendon, and M. A. Lambrix Relationships Among Caregivers' Demographic Characteristics, Social Support Ratings, and Expectations of Computer-Mediated Support Groups Journal of Applied Gerontology, February 1, 2007; 26(1): 58 - 77. [Abstract] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
