2. An economic theory of sharing

A neoclassical economic theory of sharing is perhaps best articulated in James Buchanan's Reference Buchanan1965 paper, ‘An economic theory of clubs’. Buchanan opens his paper by highlighting the pervasiveness of what Yochai Benkler (Reference Benkler2004) calls ‘shareable goods’:

Buchanan ultimately focuses on how formal clubs help people share goods like golf courses. He argues that clubs will accept new members until the cost of sharing the good with the marginal member exceeds the benefit of sharing the expense with the marginal member (Buchanan, Reference Buchanan1965: 5). In this model, the market guides individuals to share non-rival goods efficiently through clubs. Although Buchanan focuses on the efficacy of private clubs, his theory outlines the economic rationale for all sharing institutions.

While Buchanan's model provides a neoclassical argument for why we have institutions for sharing non-rival goods, Gary Becker's Reference Becker1965 paper ‘A theory on the allocation of time’ explains why people with higher incomes may be less likely to use these institutions. In Becker's model, households are ‘small factories’ that combine labour and intermediate goods, such as food and appliances, to produce final commodities, such as meals. Everything else being equal, individuals with higher wages will use production methods that are more goods-intensive and less time-intensive (Becker, Reference Becker1965: 513). Becker argues that this model explains lifestyles in the US.

Extrapolating Becker's argument to sharing, income growth will tend to reduce sharing, as long as sharing institutions are more time-intensive than private provisioning.

An array of cross-sectional, longitudinal, and qualitative evidence supports this theory that higher incomes cause lower levels of sharing. For example, Alejandrina Salcedo, Todd Schoellman and Michèle Tertilt's paper ‘Families as roommates’ models households as clubs, in which people live together only if the benefit of sharing the expense of household public goods outweighs the time cost of ‘forming and maintaining relationships’. As wages increase, people are less inclined to pay that cost and share households. Salcedo et al. (Reference Salcedo, Schoellman and Tertilt2012) calibrate their model with current cross-sectional data from the Consumer Expenditure Survey and argue that income growth explains 37% of the decline in the number of adults in the average household from 1850 until 2000 (Salcedo et al., Reference Salcedo, Schoellman and Tertilt2012: 153). Using a very different methodology, Peter Menzel's book Material World: A Global Family Portrait provides striking visual evidence that people share more in poor countries than rich countries. The book collects photographs of families with all their possessions outside their homes. Perhaps the most striking pattern is that people in poor countries share cramped quarters and few goods with a large number of relatives, while people in rich countries live with just a handful of family members in huge homes with enormous mountains of stuff (Menzel, Reference Menzel1994). Finally, the surveys used in this paper find that saving money is a primary motivation for sharing. The CNADS reveals that 72% of Americans agree with the statement ‘sharing saves money’. This is consistent with the NeighborGoods Survey (NGS), which finds that ‘saving money’ is a primary reason for using the sharing platform NeighborGoods, just below ‘reducing waste’ but well above ‘helping others’, ‘building community’ or ‘meeting people’.

Even if sharing declines as income rises, it is possible that new institutions could bring forth a new era of sharing. Yochai Benkler (Reference Benkler2004) vigorously challenges the idea that sharing is passé, but his thesis that sharing is emerging in the 21^st century as a growing ‘modality of economic production’ is compatible with the neoclassical model articulated by Buchanan (Reference Buchanan1965) and Becker (Reference Becker1965). Benkler argues that new decentralized institutions for sharing goods are likely to become increasingly important, because they reduce participation costs in terms of time and commitment. He specifically contrasts his notion of decentralized sharing with Elinor Ostrom's (Reference Ostrom1990) concept of community governance.

While traditional institutions for sharing goods depend heavily on recurrent interactions, Benkler stresses the ‘fluidity’ of participation in new institutions for sharing goods. Although he acknowledges that these new forms of cooperation may be less appealing to ‘communitarians’ who prefer the forms of cooperation found on hippie communes or in Amish communities, Benkler contends that this fluidity makes these decentralized institutions attractive to ‘many more people’ so that they are ‘likely to be more economically effective and efficient on a larger scale’ (Benkler, Reference Benkler2004: 343). Online platforms connect people to much larger networks of people and can potentially help people exploit many more opportunities to share non-rival goods. The sharing economy may lead to a resurgence of sharing among ordinary people in wealthy countries by changing how they share.

Benkler's optimism is consistent with neoclassical models if new sharing institutions provide more convenient ways of sharing non-rival goods. When sharing is time-intensive, the cost of sharing is greater for high wage earners, and the wealthy will tend to use sharing institutions less than the poor. By reducing the time it takes to share items, online platforms make sharing more attractive to people of all incomes, but the effect is greatest for people with a high opportunity cost to time. A decrease in the time cost of sharing may make sharing less sensitive to income and more compatible with continued economic growth. The sharing economy may also change the relationship between income and sharing because purchasing second-hand goods on Craigslist or travelling on Airbnb or Couchsurfing does not create long-term commitments between participants. Although users can and do convert some pleasant interactions into lasting friendships (Lauterbach et al., Reference Lauterbach, Truong, Shah and Adamic2009), they can also avoid forming relationships and warn other group members about unpleasant experiences. The arm's-length nature of social interactions in the sharing economy may significantly reduce the cost of sharing, particularly for people with high incomes who worry that traditional institutions create long-term responsibilities. By reducing the cost of coordinating beneficial peer-to-peer interactions, new institutions can potentially facilitate greater sharing in affluent societies.

Of course, there are reasons to be optimistic about the future of sharing without maintaining that online platforms will facilitate sharing among high-income people. For example, people may share more as sticky norms and endogenous preferences evolve to harness these new technologies (Fremstad, Reference Fremstad2016). The rapid innovation of new sharing institutions may also lead to success. Elinor Ostrom and Xavier Basurto's (Reference Ostrom and Basurto2011) analysis highlights that some institutions, including many traditional sharing institutions, evolve unconsciously, while others, including many new sharing platforms, consciously adapt to attract users. New institutions learn from the variation in rules across platforms, the publicness of these experiments, and the frequency of the problems that arise (Ostrom and Basurto, Reference Ostrom and Basurto2011). While there is early evidence that online platforms may increase some forms of sharing (Fremstad, Reference Fremstad2017), the argument is largely theoretical in nature.

This paper tests the argument that new sharing institutions will succeed in part because they appeal to people of all incomes. I shed light on the future of sharing by combining new data on who uses traditional and new sharing institutions with the ‘Varian Rule’ that ‘a simple way to forecast the future is to look at what rich people have today’ (Krugman, Reference Krugman2015; Varian, Reference Varian2011). While this analysis provides an imperfect view of the future, it has an empirical grounding. If the rich as well as the poor use online platforms to share goods, then the Varian Rule suggests that these institutions may remain economically important in the face of continued material growth. There is some preliminary evidence that new institutions do attract high wage earners. For example, users of the online encyclopaedia Wikipedia are disproportionately high-income (Zickuhr and Rainie, Reference Zickuhr and Rainie2011), and contributors are disproportionately well-educated, perhaps due to the ‘ease’ of making contributions (Safner, Reference Safner2016). While the approach in this paper is not fool-proof, it provides the first comprehensive empirical analysis of who uses a variety of sharing institutions, which can deepen our understanding of sharing today and may inform our expectations about the future.

3. Data

This analysis of income and sharing uses data from the 2014 American Community Survey, a 2014 Center for a New American Dream Survey, and my own 2013 survey of NeighborGoods users. Each source provides data on the use of multiple methods of sharing. My six traditional institutions for sharing are carpooling, using thrift stores and garage sales, living with a non-relative (besides a partner), living in a multi-generational household, and sharing items with relatives, friends, and neighbours. My four new institutions for sharing are exchanging goods on used merchandise websites like Craigslist, using peer-to-peer lodging platforms like Airbnb or Couchsurfing, borrowing cars on Zipcar or RelayRides, and participating in bike-sharing programmes like New York City's Citibike. In general, traditional forms of sharing are more time-intensive and rely on stronger social ties, while new forms of sharing are less time-intensive and function among loosely connected individuals. While all the traditional institutions operate without the internet, all the new institutions are built around it.

My first source of data is the 2014 American Community Survey (ACS). I use the Public Use Microdata, which provides information on nearly 2.3 million US adults. The ACS asks workers how they usually commuted to work in the previous week. Table 1 shows that 9% of workers report carpooling to work, about double the number who take mass transportation. The ACS also collects data on whether respondents live with non-relatives or live in multi-generational households, both of which may depend partly on economic factors. In the US, 9% of adults live with a non-relative besides an unmarried partner.Footnote ² The ACS also reveals that 8% of adults live in a multi-generational households, and there is very little overlap between adults living with non-relatives and those living in multi-generational households.

Table 1.

Use of various institutions for sharing goods

Note: All summary statistics use population weights, except data from the NGS.

My second source of data is a 2014 CNADS, which provides a unique look at how 1,646 Americans use one traditional institution and four new institutions for sharing goods. I use CNADS data on how often people use thrift stores and garage sales, online second-hand markets like Craigslist, peer-to-peer lodging platforms like Airbnb and Couchsurfing, car-sharing services like Zipcar and RelayRides and bike-sharing services like Citibike. Shopping at thrift stores and garage sales is quite time-intensive, and I view these as a traditional institution for sharing goods. On the other hand, the four new institutions are specifically designed to reduce the amount of time it takes to exchange used goods, find a place to spend the night or borrow a car or bike. Whether they facilitate peer-to-peer transactions, like Craigslist, Airbnb, Couchsurfing, and RelayRides, or more centralized forms of sharing, like Zipcar and Citibike, these new platforms all work to promote cooperation among loosely connected networks of people. For example, unlike car rental companies that clean and inspect their vehicles each time they are returned, Zipcar asks its customers to keep them clean, transport pets in carriers, and promptly report damage, dirtiness, or low fuel (Zipcar, 2015).

My third data source is my own online survey of 298 NeighborGoods users, which provides information on how often people borrow and lend goods across households. NeighborGoods is an online platform for sharing household goods, like ladders and power drills. This paper simply uses the NGS for information on how often people informally share goods with others off the platform. The NGS suggests that 38% of adults lend out at least one item a month and 33% of adults borrow at least one item a month.

Table 1 lists the fraction of adults who report using each institution of sharing goods analysed in this paper. The ACS asks all respondents who they live with, and it asks workers how they usually commuted to work in the previous week. The NGS and CNADS ask respondents how frequently they use a given method of sharing goods. I construct binary variables specifying whether a given individual reports using a method of sharing at least once a month or at least once a year, depending on how commonplace it is. For example, I consider the probability that someone uses Craigslist at least monthly, but I consider the probability that someone uses Airbnb or Couchsurfing at least yearly. This reflects the fact that Craigslist's for sale section has changed how people acquire stuff day-to-day, while Airbnb and Couchsurfing has changed how people find lodging on their less frequent trips and vacations.

The ACS and CNADS are nationally representative surveys, which I analyse using the appropriate population weights. The NGS provides information on a smaller sample of NeighborGoods users. Table 2 shows that the sample in the NGS is somewhat more male and somewhat younger than the adult American population as a whole. Despite these discrepancies, respondents report lending items to strangers at levels very similar to those found in the General Social Survey (Fremstad, Reference Fremstad2016), so the patterns of sharing reported in the NGS may nevertheless be similar to the American population. The largest difference between these surveys is in household income. My measure of household income also varies across surveys. The ACS provides a continuous income measure with limited top-coding.Footnote ³ The CNADS and NGS ask respondents to select their household income from a list of categories. CNADS respondents choose among seven income brackets, and NGS respondents choose among five income categories. I assume that the household income of CNADS and NGS respondents is equal to the mean income in each bracket in the ACS. Table 2 shows that mean income in the CNADS is lower than in the ACS, but mean income in the NGS is higher than in the ACS. Since I focus on the relationship between income and sharing within each sample, differences across samples do not pose a serious problem for my analysis.

Table 2.

Comparison of common variables

Note: Population means with standard deviations in parentheses.

4. Income and sharing

Using the data described above, I estimate the effect of household income on the likelihood that an American adult regularly uses each traditional and new institution for sharing goods using the following linear probability model:

(1) $$\begin{eqnarray} {\rm{Pr}}\left( {{\it sharing}{_i}} \right) &=& \sum _{j = 1}^7{{\rm{\alpha }}_j}{\it hh.inc_i} + {{\rm{\beta }}_1}{\it loghh.size_i} + {{\rm{\beta }}_2} {\it age_i} + {{\rm{\beta }}_3}{\it female_i}\nonumber\\ && + {{\rm{\beta }}_4}{\it white_i} + {{\rm{\beta }}_5}{\it bachelors_i} + {{\rm{\varepsilon }}_i} \end{eqnarray}$$

This equation estimates the probability than an individual makes use of a particular institution for sharing goods as a linear function of household income, log household size, age, whether the respondent is female, whether the respondent is white, and whether the respondent has at least a bachelor's degree.Footnote ⁴ To highlight the relationship between income and sharing, I suppress the constant and centre my control variables at their respective means. As a result, estimates of α_j can be interpreted as the probability that an individual in each income bracket makes use of an institution for sharing goods, after controlling for household size, age, gender, race and education. Using data on the use of ten sharing institutions, I test the hypothesis that sharing declines with income.

Traditional institutions for sharing

I first analyse how income affects the likelihood that individuals make use of traditional institutions for sharing goods, from carpools, thrift stores and garage sales, to living with non-relatives, living in multi-generational households, and sharing items across households. My estimates of Specification (1), which are shown in Table 3, provide no evidence that age, gender, race, or education has a consistent impact on individuals’ propensity to use traditional institutions for sharing goods. However, I do find that individuals in larger households are more likely to use all six traditional institutions for sharing goods. This may be because living with others is itself a form of sharing (Salcedo et al., Reference Salcedo, Schoellman and Tertilt2012) and individuals who share their homes with ‘roommates’ are also more inclined to engage in other forms of sharing. It may also simply reflect the fact that, controlling for household income, people living in larger households have lower per capita incomes and are less likely to use traditional sharing institutions.

Table 3.

Inverse relationship between income and the use of traditional sharing institutions

Note: Results from a linear probability model. I repress the constant and center the control variables at their respective means to ease the interpretation of household income (brackets 1–7). Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1.

Figure 1 illustrates the relationship between income and the use of six traditional institutions for sharing goods. The bar graphs show my estimate of the likelihood that a person in each household income bracket uses a particular institution for sharing goods, after controlling for household size, age, gender, race and education. The whiskers depict the 95% confidence intervals of my estimates. The first panel clearly shows that adults with higher household incomes are less likely to carpool to work. About 10% of workers in the poorest households carpool to work compared to only about 6% of workers in the richest households, and these differences are statistically significant. This result is consistent with the economic model of sharing to the extent that carpooling is an institution that allows users to substitute time for money, making it particularly attractive to low-income people. The point estimates suggest that affluent Americans are 60% as likely to carpool as low-income Americans.

Figure 1.

Inverse relationship between income and use of traditional sharing institutions

Note: This figure illustrates estimates from a linear probability model on the likelihood that someone in each category of household income regularly using a particular institution for sharing goods, after controlling for household size, age, gender, race, and education. The bar graph illustrates point estimates and the lines mark the 95 percent confidence intervals.

The remaining panels in Figure 1 present evidence that this inverse relationship between income and sharing holds across all six traditional institutions. The effect of income on sharing is more statistically significant in the forms of sharing observed in the ACS, where I can analyse over one million respondents, but the magnitude of the effect is similar across all forms of sharing. The second panel suggests that people in the highest income bracket are about 71% as likely as people in the lowest income bracket to make use of thrift stores or garage sales on a monthly basis. Although the CNADS is smaller than the ACS and the confidence intervals of my estimates are wider, the difference between low-income households and high-income households remains statistically significant. The ACS also provides strong evidence that poorer individuals are more likely to live with non-relatives or with multiple generations. Compared to people in the lowest income bracket, the point estimates in the third and fourth panels show that people in the highest income bracket are 69% as likely to live with a non-relative and 68% as likely to live in a multi-generational household, after controlling for household size and other variables. These results suggest that the decision to share a home with a non-relative or with one's extended family is partly an economic decision, consistent with Salcedo et al.’s (Reference Salcedo, Schoellman and Tertilt2012) model. There is also evidence in the NGS that people with higher household incomes are less likely to share items with people who do not live with them. My point estimates suggest that, compared to people living in households earning less than $40,000 a year, people living in in households earning over $100,000 a year are 74% as likely to lend goods and 59% as likely to borrow goods each month. Due to the small sample size of the NGS, though, these differences are not very statistically significant. Taken together, Figure 1 suggests that high incomes reduce the likelihood of using traditional institutions for sharing.

New institutions for sharing

Next, I investigate the relationship between income and the use of new institutions of sharing goods, such as Craigslist, Airbnb and Couchsurfing, Zipcar and RelayRides, and Citibike. My estimates of Specification (1) are shown in Table 4. Again, there is no indication in my results that gender, race, or education have consistent effects on the propensities to share. However, these results do show that young people are more likely to use these new platforms than their older counterparts. The results suggest that every 10 years in age reduces the likelihood that an individual regularly uses a website like Craigslist by about 15%. Assuming that young people today continue to use sharing platforms at similar rates in the future, this suggests that new institutions for sharing goods will become relatively more important as younger cohorts replace their older counterparts.

Table 4.

No relationship between income and the use of new sharing institutions without accounting for differential internet access

Note: Results from a linear probability model. I repress the constant and center the control variables at their respective means to ease the interpretation of household income (brackets 1–7). Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1.

In the long run, the future of sharing probably depends on whether new institutions can appeal to high-income people. Figure 2 reveals little relationship between income and the use of these new methods of sharing goods. The point estimates suggest that people in the richest households are sometimes more likely and sometimes less likely to use new institutions for sharing goods as people living in the poorest households. The confidence intervals of my estimates of the impact of household income on an individual's propensity to use new institutions for sharing are considerably wider than they are for most of the traditional forms of sharing observed in the ACS, because the CNADS has a much smaller sample size. The combination of similar point estimates and the large standard errors means that the linear probability model provides no evidence of statistically significant differences in the use of new institutions for sharing across income. The reason for this could be that new institutions for sharing are not as time-intensive as traditional institutions, making them attractive to both low- and high-income people, consistent with Becker's (Reference Becker1965) model. Sharing goods among fluid networks of loosely connected individuals might also be attractive to a wider range of people, as Benkler (Reference Benkler2004) argues.

Figure 2.

No relationship between income and the use of new sharing institutions without accounting for differential internet access

Note: This figure illustrates estimates from a linear probability model on the likelihood that someone in each category of household income regularly using a particular institution for sharing goods, after controlling for household size, age, gender, race, and education. The bar graph illustrates point estimates and the lines mark the 95 percent confidence intervals.

While there is clear evidence that traditional forms of sharing decline with income, it appears that Americans with higher incomes are no less likely to use new services like Craigslist, Airbnb, Zipcar, or Citibike. The reason for this may be that these forms of sharing are less time-intensive than more traditional forms of sharing. These preliminary results also suggest that the sharing economy may succeed in increasing sharing levels over the long run, even in the face of continued economic growth.

Accounting for internet access

Although the above analysis suggests that high-income people are no less likely than low-income people to make use new institutions for sharing, such as Craigslist, Airbnb or Zipcar, it ignores the important fact that people need reliable internet access to effectively share goods using online platforms. If low-income people have worse access to the internet, that could explain why they appear no more likely to use traditional institutions for sharing goods than high-income people. I consider this possibility using ACS data on whether respondents live in a household that subscribes to some form of internet service, such as a cable, DSL or mobile broadband plan, or whether it accesses the internet in some other way. Using my linear probability model, I estimate the likelihood that a person in any income bracket has home internet service. As shown in Figure 3, the digital divide remains stark. Whereas 92% of people living in households earning over $150,000 have internet access, only 62% of people living in households earning under $25,000 do. Given the large sample size of the ACS, these estimates are extremely precise so that the 95% confidence intervals are barely visible in Figure 3. For example, the estimates imply that there is a 95% chance that the true percentage of people with some form of internet access among the poorest households is between 62.3 and 62.7%.

Figure 3.

Internet subscriptions and internet access across income

Unfortunately, the CNADS does not ask respondents about whether they have internet access, so I cannot simply restrict my analysis to respondents with internet access in Specification 1. Nevertheless, the ACS data on the digital divide allows me to disentangle the true relationship between income and sharing in the CNADS. I reweight the CNADS survey by multiplying the survey weight by the probability that the respondent lives in a household with internet access. Under the assumption that anyone who regularly uses any of the online sharing platforms must have internet access, I assign a weight to non-users such that the probability that a CNADS respondent in each income bracket has internet access is consistent with the probability that an ACS respondent in the same income bracket has internet access. Using these weights, I then re-estimate the effect of income on whether people use new institutions for sharing. The purpose of this exercise is to investigate the relationship between household income and the use of new institutions for sharing goods among people who can access these online platforms.

My estimates of the probability that people with internet access make use of new institutions for sharing goods across household incomes are shown in Table 5 and Figure 4. The first column of Table 5 addresses whether people use online markets for second-hand goods, such as Craigslist. Among people with internet access, the point estimate suggests that 47% of individuals living in the poorest households use websites like Craigslist at least monthly, compared to 36% of individuals living in the richest households. Using the adjusted weights, my point estimates all suggest that affluent people are less likely to use new sharing institutions than the poor: people in the highest income bracket are 78% as likely to use websites like Craigslist, 77% as likely to use platforms like Airbnb or Couchsurfing, 47% as likely to use services like Zipcar and RelayRides and 40% as likely to use bike-sharing systems. These estimates are not very precise, and the 95% confidence intervals for these two income categories overlap much more than the confidence intervals presented in Figure 1. In the following section I address the statistical robustness of this negative relationship between income and new forms of sharing.

Table 5.

Inverse relationship between income and the use of new sharing institutions, when accounting for differential internet access

Note: Results from a linear probability model. In this model, I weight the CNADS data by the probability that each respondent has internet access, as described in the text. I repress the constant and center the control variables at their respective means to ease the interpretation of household income (brackets 1–7). Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1.

Figure 4.

Inverse relationship between income and use of new sharing institutions when accounting for differential internet access

Note: This figure illustrates estimates from a linear probability model on the likelihood that someone with internet access in each category of household income regularly using a particular institution for sharing goods, after controlling for household size, age, gender, and race. The bar graph illustrates point estimates and the lines mark the 95 percent confidence intervals.

5. Robustness

To my knowledge, these are the first published estimates showing that people with high incomes are less likely than people with low incomes to use both traditional and new institutions for sharing goods. Although the differences in the probability that individuals use institutions are not always statistically significant across income categories, my finding that high-income people are less likely to use all sharing institutions is reasonably robust.

First, the magnitude of the estimates is quite similar across all ten institutions for sharing goods. The point estimates in Table 3 suggest that Americans in the highest income bracket are about 67% as likely as people in the lowest income bracket to use traditional sharing institutions. Meanwhile, the point estimates in Table 5 suggest that, after accounting for differences in internet access, rich people are about 61% as likely as poor people to use new sharing institutions. These differences in use between high- and low-income people are much more statistically significant for forms of sharing that are observed in the ACS, but this mostly reflects the fact that the ACS covers 1,000 times as many people. As more data become available on who uses new online platforms, researchers will be able to estimate the relationship between income and sharing with greater precision, but current data suggest that the relationship is similar across all forms of sharing.

Second, although the differences between how much people in the highest and lowest income categories share goods are not always statistically significant, this is partly due to the fact that I divide my sample into five to seven income categories. As a robustness check, I simply divide the CNADS into two groups of roughly equal sizes: people with household incomes below $50,000 and people with household incomes above $50,000. I find statistically significant evidence that low-income people are more likely to use three of the new institutions for sharing. Specifically, I reject the proposition that the rich and poor are equally likely to use websites like Craigslist at the 1% level, Airbnb and Couchsurfing at the 10% level and Zipcar and RelayRides at the 5% level. This test does not show a statistically significant difference in use of bike-sharing services, which Figure 4 suggests are quite popular among people with incomes between $50,000 and $150,000, but much less popular among people with incomes above $150,000. The inverse relationship between income and sharing is clearest on Craigslist, which is the most popular platform analysed by CNADS.

Third, an alternative specification suggests that the negative relationship between income and sharing is statistically significant for five traditional institutions and two new institutions. Section 4 presents estimates for a linear probability model, but I reach similar conclusions when I use a logit model to estimate the likelihood that a person regularly uses a sharing institution as a function of log household income.Footnote ⁵ In the case of traditional institutions for sharing, I find strong evidence (statistically significant at the 1% level) that people in more affluent households are less likely to carpool, shop at thrift stores or garage sales, and live with non-relatives, and weaker evidence (statistically significant at the 10% level) that they are less likely to borrow and lend items across households. When I ignore differences in internet access, I find no statistically significant relationship between income and the use of new institutions for sharing. However, once I reweight the data to reflect the probability that any individual has home internet access, all my point estimates suggest that people in affluent households are less likely to share, including strong evidence (statistically significant at the 1% level) that the affluent are less likely to use Craigslist and weaker evidence (statistically significant at the 10% level) that the affluent are less likely to use Zipcar or RelayRides. This paper focuses on the results from my linear probability model, because they are simpler to interpret, but the key results are consistent with the results from a logit model. The relationship between income and sharing should be re-examined as more data become available, but the evidence presented in this paper suggests that people with higher incomes are less likely to use both traditional and new institutions for sharing goods.