Extracted main text — title through conclusion, appendix excluded. This is what our citation measures are computed over, published so the extraction can be checked by eye.
30,046 characters · 10 sections · 16 citation commands
Online Red Packets: A Large-scale Empirical Study of Gift Giving on WeChat
Online red packet (formerly called “Lucky Money” or “Red Envelopes”) was first released in 2014 as a new feature on WeChat, the largest social messaging platform in China. This feature enables people to send red packets to individual friends individually or randomly split a red packet among a limited number of people in a group on a first-come, first-serve basis. WeChat red packets have become viral since 2015: there were over 100 million participants during the 2016 Mid-autumn Festival alone, and the number of red packets sent over WeChat on Lunar New Year's Eve 2017 was 14.2 billion. Now, to facilitate paricipation by users overseas, it is possible for someone to send red packets by linking her PayPal account to WeChat. \footnote{The Economist offers an introduction to WeChat and red packets (\href{https://www.economist.com/news/business/21703428-chinas-wechat-shows-way-social-medias-future-wechats-world}{https://www.economist.com/news/business/21703428-chinas-wechat-shows-way-social-medias-future-wechats-world}). }
The red packet phenomenon is an instance of gift giving, which is a pervasive behavior in human society that ranges from giving Christmas gifts to purchasing from a wedding registry cheal2015gift,mauss2000gift,sherry1983gift,caplow1984rule,akerlof1984gift,schwartz1967social,bendapudi1996enhancing. However, gift giving is governed almost entirely by unwritten rules. Such unwritten rules have attracted interest from a wide range of disciplines, including anthropology, economics, sociology, psychology, and marketing cheal2015gift,mauss2000gift,sherry1983gift,caplow1984rule,bendapudi1996enhancing. Findings indicate that the motivations for gift giving include altruism andreoni2003charitable,midlarsky1989generous, reciprocity komter1997gift,komter1996reciprocity, and many other factors such as prestige harbaugh1998donations, empathy berger1962conditioning, etc. For example, the red packets between friends are mainly reciprocal and it is considered as “Renqing (favor)” in Chinese culture chan2003art. By contrast, sending red packets among family members is expected to enhance family bonds is usually considered non-reciprocal.
Our contribution to the literature is twofold. First, we study a new type of online red packets, the “group red packet”. In contrast with traditional red packets exchanged among family members and friends, this feature allows individuals to send red packets to a group of people by randomly splitting a red packet among a limited, user-determined number of people on a first-come, first-served basis. Moreover, online platforms provide a convenient and informal channel for gift giving, which may bring fundamental changes to the age-old practice of gift giving. Second, most studies in the literature use small samples, including studies of online gifts taylor2002age,suhonen2010everyday. To the best of our knowledge, this is the first study to quantitatively analyze gift exchange using a dataset on millions of users. In addition, considering a different setting than the fruitful existing empirical works focusing on the West andreoni2003charitable,kottasz2004differences,mesch2011gender,midlarsky1989generous,hansler1989geo,radley1995charitable,mears1992understanding,amato1987family,jones1991charitable,berg1995trust,komter1997gift,komter1996reciprocity,wang1997does,gneezy2006putting,plickert2007s,falk2007gift, our study includes a large sample of the population in contemporary China and sheds light on gift giving in a representative Eastern country.
In this paper, we study this new type of gift giving behavior in three respects. First, we provide an overview of the cash flow patterns among users with different demographic and geographic backgrounds. If gift exchanges were completely reciprocal, the cash flows among users from different groups would be approximately zero. However, our results reveal intriguing patterns among different demographic and geographic groups. For example, we find that users from southern provinces are inclined to export cash to their northern counterparts.
Second, we study how demographic factors such as gender and age relate to sending red packets. In particular, to distinguish the motivations for sending red packets (altruism vs. reciprocity), we define spontaneous and reciprocal red packets and identify their corresponding distinctive patterns. We find that males, older people, southerners, and people with more friends in the group tend to send more spontaneous red packets, e.g., they are more likely to be the first one to send a red packet in a group, while red packets from females, younger people, northerners, and people with more friends in the group are more reciprocal. These differences can be explained by many aspects of Chinese culture, including “mianzi”, i.e. the Chinese cultural value of interpersonal dignity or prestige.
Finally, we examine the casual effect of online red packets on group dynamics. Applying propensity score matching to data on approximately seven million red packets, we conduct a quasi-experiment to study the causal effects of the cash amount in red packets. We show that increasing the amount of money in red packets can stimulate reciprocal red packets and, consequently, increase the total amount of red packets sent by members of a group. We also find that increasing the amount of money in red packets can provide the senders with new friends. Based on these findings, we shed light on the benefits of online group red packets for both senders and all group members and improve our understanding of the motivation for sending red packets.
We measure the directions of cash flows between users with different demographics. In particular, we are interested in examining whether, on net, the group red packets are sent from those from more-developed provinces to those from less-developed provinces.
Figure (ref) (a) presents the net cash flow in Mainland China.\footnote{For easy visualization, we filter pairs of provinces where the total cash flow is under 1,000 RMB in the dataset.} First, we observe more net cash flow from the east coast to the west, or from the south to the north. This suggests that the direction of the cash flow might be significantly affected by the GDP in a region. Second, the magnitude of cash flow varies and the largest cash flow is from Guangdong to Hunan. We conjecture that this is driven by the large number of migrants from Hunan to Guangdong.
Furthermore, we are interested in understanding the relationship between the magnitude of cash flows and the geographic distance between two regions. We selected (1) Beijing, which is the capital of China and a municipality in the north with a large number of migrants, (2) Heilongjiang, which is a province in the northeast that has been experiencing significant economic difficulties and labor force outflows, and (3) Guangdong, which is the province with the highest GDP, population, and number of migrants.
We find that the net cash flows from Heilongjiang to other provinces are almost always negative and the figures from Guangdong to other regions are almost always positive. Furthermore, the cash flows for both provinces are mainly with neighboring regions, and geographic distance tends to have a negative impact on the level of cash flows. In contrast, we do not find a consistent direction of cash flows for Beijing. On the one hand, the average standard of living in Beijing is higher than in most other provinces, leading to substantial cash flows out of this region; on the other hand, there are millions of college students in Beijing, and this may lead to large cash flows into the region. The cash flows between two arbitrary provinces and the net amount of incoming cash for each province are presented in SI.
To validate the aforementioned factors affecting the interprovincial cash flows, we predict the net cash flow between two provinces by the differences in GDP and GDP per capita (PPP) and the distance between their capitals. We find that a one-thousand-RMB increase in the difference in PPP is associated with a $9.02$ increase in the net cash flow ($p<0.01$), while a one-kilometer increase in distance is associated with a $0.57$ decrease in the net cash flow ($p<0.01$), while the results for GDP are not significant. The difference in PPP measures the difference in standards of living between provinces, and these results indicate that it is more likely to see a net cash flow from a more-developed to a less-developed province. We explain the effect of the distance by simple geographic proximity: nearby provinces have more interactions and migration between them.
Additionally, we also examine the overall cash flows between people of different ages and find that, generally, the net cash flow is from older to younger people. This is consistent with social norms in China, i.e., older people give red packets to younger people, especially to children and the unmarried. In terms of gender, we find that, in general, the direction of the net cash flow is from males to females. Comparing four types of cash flow (male to male, male to female, female to male, and female to female), we find that male to male represents the largest proportion (36%). Details on cash flows among age and gender groups are stated in SI.
In addition to overall cash flows, we examine the motivation for sending red packets by distinguishing two types of red packets and the effect of different factors on the likelihood of sending red packets.
Reciprocity is considered an important motive for gift givingkomter1996reciprocity. We distinguish a particular type of red packets: reciprocal red packets. If a person receives a red packet and then quickly sends a new red packet, we consider the new red packet to be reciprocal. We also define spontaneous red packets, which are initiated without the sender having recently received a red packet, and thus, these are unlikely to be triggered by reciprocity. The formal definitions are as follows:
We define propensity to send given feature $F$ (see Methods). Fig. (ref) presents the average propensity score for key features across all groups. We then define the relative propensity to send each particular type of red packets given a feature $F$. Specifically, we subtract the propensity to send any red packet from the propensity to send a spontaneous/reciprocal red packet. The results are shown in Fig. (ref). Details of definition of the variables can be found in SI.
We have the following observations based on Figures (ref) and (ref). Please refer to SI for the definition of related variables.
We have similar analysis on receivers, which can be found in SI.
Although reciprocity is a widely accepted motivation for gift-giving, it cannot explain the motivations for all red packets. For example, sending a spontaneous red packet is unlikely to be driven by pure reciprocity. herefore, in this section, we attempt to measure the causal effect of spontaneous red packets on the group, as well as the red packet's sender, by applying propensity score matching to examine the possible benefit that red packets bring to groups.
Since it is difficult to conduct counter-factual analysis such as “what if this red packet had not been sent”, we choose to match and compare the spontaneous red packets with large amounts of money compared to those with small amounts of money. Doing so allows us to determine whether a marginal increase in the money in a red packet can have significant impacts on group activities. Therefore, our treatment dummy is coded as one when the cash amount of a spontaneous red packet $\geq$ 100 RMB.\footnote{We also used other thresholds, i.e., 50 and 150 RMB, and find consistent results.}, zero otherwise. In total, we have 54,150 red packets in the treatment group and 6,133,489 in the control group. Table (ref) reports the regression results. We also control for the number of splits allowed for each red packet (RPnum), and group fixed effects, including at the individual and seasonal levels.
We define a session as a sequence of red packets in a group where the interval between two consecutive red packets is shorter than $\tau$. $\tau=30\text{min}$, and we examine the effect of spontaneous red packets on the session. The dependent variables are (1) the total amount of money available to following users in this session (session money); (2) the number of following users who send red packets in the session (number of following senders, or \#FS); (3) the number of following red packets in the session (\#FRP); and (4) the number of new in-group friends that the sender has within 1 day after a spontaneous red packet is sent (\#Senders' new friends).
Compared to prior studies that match on individual-level data aral2009distinguishing, we further utilize group features, e.g., the size of the group, and seasonality features, e.g., whether the red packet is sent during the spring festival, to understand the causal effect of the amount of cash in red packets. All 54,150 treatment observations are successfully matched. The standardized differences of all covariates are below 0.1, which indicates insignificant imbalance between the treatment and control groups normand2001validating.
First, compared to small red packets, on average, large red packets have 0.295 more group members following and generate 0.415 more following red packets (Columns 1 and 2). Additionally, red packets sent on special days, such as festivals and weekends, generate more following users and following red packets. Interestingly, those sent by female users and by people with lower in-group degree also have more following users and more following red packets.
Second, we examine the impact on the following session money (Column 3). We find that a large red packet, on average, increases 52.632 RMB in session money. This indicates that a large cash amount may significantly benefit other group members. Therefore, they might be more likely to reciprocate by sending red packets to the group.
Finally, we analyze the number of new friends that a group member garners by sending a large amount of red packets. On average, a large amount of red packets attracts 0.009 new friends to the sender. This indicates that red packets can help senders to attract new friends.
In summary, we have presented the macro- and micro-level patterns in red packet sending and discussed the motivations for and benefits of sending red packets.
Studying the interprovincial cash flows improves our understanding of human mobility. The patterns that we found are consistent with a gravity model: the interactions between two areas, for example in the form of immigration and trade, are proportional to the “masses” in these two areas (population, economic development) and inversely proportional to the distance between themlewer2008gravity; similarly, cash flows between provinces are dependent on both living standards (PPP) and distance. We conjecture that the trends could be driven by immigration: people move to nearby and more developed provinces and maintain their social bonds with families and friends in their hometowns.
Our micro-level results add new dimensions to our understanding of gift giving. For example, we find that men are generally more inclined to send red packets and that red packets from women are more likely to be reciprocal than spontaneous. Most existing studies on gender differences in gift exchange, especially in charitable giving, state that women are more likely to donate to more charities kottasz2004differences,andreoni2003charitable,mesch2011gender. However, in the context of group red packets, men are more likely to send red packets. Rather than pure altruism, we conjecture that income inequality and a desire for prestige account for this gender difference. Although it has decreased, the gender income gap persists in China xiu2013gender, which limits the budget available to females to reward others. Another account is the desire for prestige, or “mianzi” in Chinese culture. Findings indicate that men are more inclined to seek status than women von2010men, and sending red packets, especially spontaneous red packets, provides a way to demonstrate one's generosity and wealth.
The results of differences in reciprocal and spontaneous red packets and casual effects of spontaneous red packets explain the motivations of sending red packets, and furthermore, reasons why online red packets has spread rapidly in China. We find that the increased rewards (amount of money) spontaneous red packets of increased rewards (amount of money) to a group can significant stimulate other group members to send follow-up red packets. Ultimately, spontaneous red packets generate rewards for group that exceed the packets' initial values. The explanation for this increase is likely reciprocity, that is, others have received money and feel obliged to send money in turn. Moreover, motivations of the spontaneous senders who tend to be senior or people with high in-group degree can be explained by the aforementioned benefits of sending red packets.
However, our work does suffer from limitations. First, despite the large sample size, our dataset is unable to represent the entire population of China. Although 65% people in China use WeChat, users who are active in red packets are biased toward young and middle-aged people. Moreover, we filtered out inactive groups and suspiciously highly active groups, which may also have influenced our results. Second, gender, age, location, and status differences can be confounded by unobserved factors such as income. However, because of privacy issues, we were unable to collect such information. Therefore, we are not arguing for causal effects of, for example, gender or age. Third, although we explored the motivations for sending red packets by examining the causal effects of spontaneous red packets, we do not have a comprehensive understanding of the true motivations of such sending behavior, although we propose that reciprocity and the beneficial effects on group activities could be two main reasons.
There are many promising directions for future research. For example, if we were to collect data from all message groups to which an individual belongs, we could explore her strategies for allocating money across the groups. In addition, we could also study how red packets are spread across all WeChat groups and which groups are responsible for the rapid adoption of red packets.
We first randomly sample one million WeChat groups in which at least one group red packet was sent between October 1, 2015, and February 29, 2016. To avoid the problem of data sparsity, we exclude groups with fewer than $\eta$ red packets. We further exclude extremely active groups with more than $\tau$ red packets as a group that have extremely large amounts of money in the red packets, as these might be used for online gambling, and incorporating these groups might significantly bias our results. Specifically, we empirically set $\eta = 3 \times m$ and $\tau=50 \times m$, where $m$ is the group size. In total, this selection process leaves us with 367,361 groups with 7,816,214 group members (7,380,110 unique users).
We combine three datasets for our statistical analyses, including (1) the characteristics of 367,361 WeChat groups, e.g., the number of members, total number of red packets and the total value of the red packets; (2) the characteristics of the 7,380,110 unique users in these WeChat groups, e.g., demographic variables, the number of friends and the number of WeChats that they join; and (3) propensities for red packets, such as the cash amount and the number of recipients of a particular red packet. In total, 61,501,862 red packets were sent in these WeChat groups, generating 218,301,860 recipients.
During this five-month period, the average number of red packets sent in a group is 172.88 and the average amount of money exchanged in a group is 995.78 RMB (approximately 146 USD). All details on the data and summary statistics of the variables can be found in the Data section of the SI.
We define a group $g \in G$, where $R_g$ represents number of red packets sent in group $g$. Among all group members (potential senders), there are $N_F$ group members with the feature in question and $N_{\bar{F}}$ without it. Among all red packets in that group, we can compute the frequency of red packets sent by people with feature $F$, $T_F$, and without feature $F$, $T_{\bar{F}}$. Then, for each group, we compute $\frac{T_F}{T_F+T_{\bar{F}}} - \frac{N_F}{N_F+N_{\bar{F}}}$ as the propensity to send given feature $F$ in this group.
To eliminate the influence of other factors, we extract various features for matching. For the group features (group-level fixed effect), we have the size of the group, density of the group network at the time red packet is sent, gender ratio, age entropy, and province entropy. For the sender features (individual-level fixed effect), we have the gender, age, location of the city (latitude and longitude), the number of WeChat friends the individual has (activity_1), the number of WeChat groups to which the individual belongs (activity_2), and the degree ratio ($\text{degree}/(\text{member count} - 1)$). For the seasonal features, we have whether it is during a festival (such as the spring festival period, New Year's Day, and Christmas), whether it is weekday, the period of a given day (we divide one day into six equal periods, as an integer; empirically, a higher number indicates a higher probability to send more and larger red packets), and the UNIX time stamp of the sent red packet (how many seconds from January 1, 1970, sendtime). Additionally, we have the number of parts into which the red packet was divided (RPnum). Imposing such controls can help to exclude the trivial statements such as “during festivals, people are more likely to send big red packets when people are also very active on WeChat to send red packets”.