Returns to Plastic Surgery in Marriage and Labor Markets Soohyung Lee Department of Economics and MPRC University of Maryland, College Park [email protected]

Keunkwan Ryu Department of Economics Seoul National University [email protected]

October 2009

Abstract Using a unique dataset from Korea, we find that plastic surgery significantly improves an individual’s facial attractiveness (“beauty”) and the improvement is larger for women and for those with a low initial level of beauty. We measure beauty premiums for an individual’s own income (i.e., labor market) and spousal income (i.e., marriage market) and then compute the monetary benefit of plastic surgery. We find that, on average, the benefit of plastic surgery is small compared to the surgery cost, but with an exceptionally large improvement, the cost can be recouped in three years.

JEL Classification: C90, J10, J12, J31 Keywords: beauty premium, marriage, plastic surgery, experiment

We thank Chungyong Han for his excellent research assistance. This paper was inspired by discussions with Caroline Hoxby. We have benefited from discussions with Dan Hamermesh, Melissa Kearney, Sejik Kim, Ben Malin, Muriel Niederle, Minjung Park, Dan Quint, Azeem Shaikh, Ed Vytlacil and participants at the 2009 Far East and South Asia Meeting of the Econometric Society and 2009 Seoul National University Summer Economics Conference.

1

1. Introduction Use of plastic surgery has drastically increased in recent decades and has become common in both developed and developing countries. In the U.S., the number of cosmetic procedures performed increased from 0.4 million in 1992 to 4.1 million in 2007, with expenditures rising from U.S. $927 million to more than $12 billion over the same period (American Society of Plastic Surgeons, 2008).1 The popularity of plastic surgery is limited neither to the U.S. nor to developed countries. The International Society of Aesthetic Plastic Surgery (2004), a worldwide association of plastic surgeons, shows that cosmetic procedures are widely performed in Mexico, Argentina, Spain and France, which comprise the top five countries together with the U.S. in terms of the number of procedures performed in 2004. Among the countries ranked between 6th and 15th are several developing countries such as Brazil, South Africa, Turkey, South Korea and Taiwan. This increasing use of plastic surgery may reflect the existence of beauty premiums in labor and marriage markets, which have been documented by a growing number of studies (e.g., for labor markets Hamermesh and Biddle, 1994; Pfann et al., 2000; Persico, Postlewaite and Silverman, 2004, Mobius and Rosenblat, 2006, Glied and Neidell, 2008, Case and Paxon, 2008; and for marriage markets Fisman et al., 2006, Hitsch et al., 2009, Lee, 2009, Banerjee et al., 2009). This paper examines the extent to which plastic surgery improves an individual’s beauty and thus allows him or her to enjoy monetary benefits due to the premium that the labor and marriage markets place on beauty. We obtain head-to-shoulder photographs of individuals in South Korea before they undergo plastic surgery procedures, and, for a subset of the sample individuals, additional photographs of their hypothetical look if they take the procedures. Big eyes, a high nose bridge, and an oval-shaped face define beauty for many Koreans. Surgical procedures that alter individuals’ facial features accordingly are popular. We recruit 50 reviewers in Korea to evaluate the facial attractiveness of individuals before and after the surgery. We find 1

From 1992 to 2007, the average annual growth rates of the number of procedures and total expenditures are 64 percent and 83 percent, respectively. For 1992, we compute total expenditure by multiplying the number of procedures by the average fee in 1992 as reported on the website of the American Society of Plastic Surgeons (ASPS). Whenever the ASPS reports a more detailed fee structure for a cosmetic procedure, we use the highest fee to compute the total expenditure, which will give us the lower bound for the annual average growth rate of the industry. For example, the ASPS reports that eyelid surgery costs $1,514 for both upper lids, $1,519 for both lower lids and $2,625 for all four lids, so we take $2,625 as the average cost per eyelid surgery for 1992.

2

that plastic surgery inquirers, on average, can experience an increase in our measure of facial attractiveness (“beauty index”) by 0.4 standard deviations if they have surgery. However, the improvement due to the surgery sharply diminishes as a patient’s initial beauty increases. Next, we examine the extent to which the facial attractiveness of a person affects his or her outcomes in marriage and labor markets. To do so, we use a uniquely rich dataset from a major Korean matchmaking company. The dataset provides detailed information for each individual such as income and family background as well as “facial grade.” Facial grade measures the facial attractiveness of a user, ranging from A for the most attractive to F for the least. It is assigned by the company’s staff members based on the user’s head-to-shoulder photograph. We find significant beauty premiums in both the marriage and labor markets, even after controlling for detailed information on the individual and his or her family background. However, beauty premiums are largely for those whose facial grade is A (top 7 percent of users), and there is no statistically significant difference in marriage and labor market outcomes between those with the median grade and those with the lower grade. For example, a user whose facial grade is above the median grade is more likely to earn higher annual income (by 7.7 percent for men and 6.9 percent for women) and more likely to marry a person with higher income (by 7.7 percent for men and 9.5 percent for women) than the median-grade user. Finally, we compute the expected monetary gain due to the plastic surgery procedures and compare it to the pecuniary cost. This cost-benefit analysis is made possible as we have access to head-to-shoulder photos for a subset of users of the matchmaking company, and thus, we are able to relate our measure of beauty (beauty index) to the company’s beauty measure (facial grade). We find that even under generous assumptions that enlarge the monetary returns to plastic surgery, it takes at least ten years for individuals to recoup the cost of surgery. This result follows from the earlier findings that plastic surgery mostly helps unattractive people obtain average beauty and that there is little beauty premium from having average, rather than low, beauty. Therefore, on average, the monetary benefit from plastic surgery is relatively small compared to the surgery cost. In contrast, if a surgery outcome turns out exceptionally good for an individual such that his or her facial grade is improved from the bottom facial grade to the top, he or she can recoup the cost in three years. This paper is related to three strands of literature. The first strand studies the beauty premium in labor markets. For example, in their seminal work, Hamermesh and Biddle (1994)

3

find that in the U.S. and Canada, workers of above-average facial attractiveness earn about 10 to 15 percent more income than their counterparts with below-average beauty. Persico et al. (2004), Case and Paxon (2008), and Deaton and Arora (2009) document a height premium in the U.S. and U.K. labor markets. Glied and Neidell (2008) study the beauty premium of having healthy teeth in the U.S. labor market. Mobius and Rosenblat (2006) use an experiment to show the existence of a beauty premium in the labor market. We find a significant beauty premium in Korea, even after controlling for detailed characteristics of individuals, their family background and self-esteem.2 The second related literature documents people’s preferences for beauty in dating and/or marriage markets. For example, Fisman et al. (2006) and Hitsch et al. (2009 and forthcoming) show that people in the U.S. are more likely to ask beautiful people for dates. Lee (2009) and Banerjee et al. (2009), by studying the Korean and Indian marriage markets, respectively, argue that people consider beauty to be an important spousal trait. We add to this literature by estimating returns to beauty in terms of spousal income, controlling for an extensive list of individual characteristics. Third, this paper is related to studies of people’s strategic behaviors to improve marriage market outcomes. For example, Bergstrom and Bagnoli (1993) present a theoretical model in which men with high ability have an incentive to delay marriage in order to allow for a stronger signal of their “high” type. Chiappori et al. (forthcoming) and Lafortune (2009) study the returns to schooling investment in marriage markets. Related to these studies, we examine facial attractiveness as an endogenous variable that people can choose to change in order to improve their marriage market outcomes. The rest of the paper is organized as follows. Section 2 presents the institutional background and the descriptive statistics of the datasets. Section 3 examines the effect of plastic surgery on facial attractiveness. Section 4 documents the beauty premiums of having attractive facial features in marriage and labor markets. Section 5 presents a back-of-the-envelope calculation of returns to plastic surgery. Section 6 discusses the potential problems with the analysis and carries out robustness checks. Section 7 concludes. 2

Our estimates of return to beauty in Korea are comparable to, but a bit smaller than, what Hamermesh and Biddle (1994) find in the U.S and Canadian markets. Although the difference may have many explanations, we think that the low level of returns to beauty may be due to the narrow dispersion of income arising from the fact that workers in our sample are rather homogeneous: they are young (the median age is 30 for women and 33 for men), mostly college educated, and employed in white-collar jobs (instead of self-employed). 4

2. Institutional Background and Data 2.1. Institutional Background In Korea, plastic surgery has been popular and growing rapidly. On the supply side, the number of surgeons specializing in plastic surgery increased from 67 in 1990 to 926 in 2000. This steep increase corresponds to an annual average growth rate of 128 percent, whereas the growth rate of the overall number of doctors in Korea is less than 1 percent per year (Lim, 2002). On the demand side, a survey conducted by the Korean Ministry of Health, Welfare, and Family Affairs (2004) finds that among 1,565 female college students, 52.5 percent of the survey participants had at least one surgical or non-surgical plastic procedure. ARA Consulting (2009, cited in Fackler, 2009) estimates that about 30 percent of women aged 20 to 50 had plastic surgery procedures. Other sources confirm the findings from these surveys (see Table 1). For many Koreans, wide eyes, high nose bridges, and oval-shaped faces (and additionally, for women, white skin) define beauty. This is illustrated by Figures 1 and 2. Figure 1 features pictures of a hypothetical Korean man and a woman synthesized from photos of 15-19 popular celebrities considered “beautiful,” whereas Figure 2 features pictures of a hypothetical average Korean man and a woman (Rhee, 2006; Rhee and Koo, 2007; Cho, 2007). Comparing Figures 1 and 2, we notice that the “beautiful” Korean man and woman have wider eyes and higher nose bridges. Column (1) of Table 2 shows the five most popular surgical and nonsurgical plastic surgery procedures in Korea. The popularity of these procedures reflects the notable differences between the average Korean and his or her ideal counterpart. In Korea, “Asian eyelid surgery” is the most popular, followed by laser mole removal and nose augmentation. Asian eyelid surgery creates an artificial crease on a person’s upper eyelids (casually referred to as “double-fold eyelid”) to make eyes rounder, wider, and bigger when a person opens his or her eyes. Figure 3.1 compares a patient’s eye before and after the Asian eyelid surgery. After the surgery, the patient has creases right above the tip of her eyelids. Note that only 28 percent of Koreans are born with the double-fold eyelids (Cho, 2007), and only half of all Asians are born with a crease in the upper eyelid, whereas most Caucasians are (Wyer, 2000).3 3

Instead, Asians have a special fold, called the epicanthic fold, which creates almond-shaped eyes (Wyer, 2000). Asian eyelid surgery is different from eyelid blepharoplasty that removes excess tissue in upper or lower eyelids and 5

Figure 1: Facial Features of “Beautiful” Koreans

SOURCE:

Rhee (2006) and Rhee and Koo (2007).

Figure 2: Facial Features of Average Koreans

SOURCE:

Cho (2006).

Laser mole removal is a non-surgical procedure for obtaining whiter and cleaner skin by removing moles, dark spots, or skin irregularities. Nose augmentation is a surgical procedure that elevates a patient’s nose bridge to “provide a natural, pleasing ridge to a flat nose” (Wyer, 2000). Nose augmentation can be combined with a surgical procedure for reshaping the tip of one’s nose to create a “sharper” image (see Figure 3.2). Nose augmentation is different from typical nose reshaping procedures, often used for Caucasians, which aim to reduce the size of the nose or to remove an unwanted hump. The use of surgical procedures to alter facial features appears to be more common in Korea than in the U.S. (see columns (2) and (3) of Table 2). According to the American Society of Plastic Surgeons (2008), 11.7 million plastic surgery procedures

is often performed for elderly people in the U.S. 6

Figure 3: Illustration of Some Plastic Surgery Procedures 3.A – Asian Eyelid Surgery

Before surgery

After surgery

3.B – Nose Reshaping Surgery

Before surgery SOURCE:

After surgery

Apgujung Seoul Plastic Surgery clinic (http://www.seoulps.co.kr/).

(0.038 procedures per capita) took place in the U.S. in 2007. The five most popular plastic surgery procedures in the U.S. are Botox injection, hyaluronic acid (for anti-aging), chemical peel, laser hair removal and microdermabrasion. Among persons in the 20 to 29 age bracket, comparable to the Korean survey respondents in terms of age, nose reshaping is the only surgical procedure in the top five list, occupying 11 percent of all plastic surgery procedures. In Korea, surgical procedures altering facial features consist of about 29 percent of all procedures as reported by the survey participants (i.e., eyelid surgery: 25.3 percent; nose augmentation: 3.6 percent).

2.2 Data Description and Empirical Strategy We use two datasets (see Table 3). The first dataset is from a Korean plastic surgery consulting company that was established in 2003. The company provides plastic surgeons in Korea with a 7

computer software program that simulates a hypothetical face of a potential client assuming that he or she undergoes certain plastic surgery procedures. Since November 2005, the company has directly provided potential patients with a hypothetical look if they undergo certain plastic surgery procedures. To receive this free online consulting from the company, one needs to post his or her head-to-shoulder photo and also needs to agree to make the before and after surgery photos available to the general public via the company’s website. Out of all the available photos from the website as of December 2008, we choose 56 individuals (“inquirer sample”) whose photo qualities in terms of resolution and size are comparable to those of the photos in the second dataset, which is described in the next paragraph. For each individual, we know what plastic surgery procedures he or she has expressed interest in, the corresponding cost, his or her current head-to-shoulder photo and a hypothetical look after the plastic surgeries. Potential concerns regarding using hypothetical looks instead of actual results are investigated in Section 6.1. The second dataset is from a major Korean matchmaking company, which helps its users find a spouse from among other users of the opposite sex (see the Appendix for a more detailed description of this dataset). The dataset includes more than 20,000 users who subscribed to the company’s matchmaking services. There are two types of users in our sample: “regular members” and “temporary online members.” A regular member pays about $900 for a one-year membership and needs to submit legal documents to verify key information about his or her traits. On the other hand, a temporary online member pays $50 to participate in a special online dating event and does not need to submit legal documents for verification purposes. In addition, regular members are assigned facial grades by the company based on head-to-shoulder photos, whereas temporary online members self report their facial grades. We obtain 140 head-toshoulder photos of users (79 of them are regular members) and detailed information about each user, such as his or her socio-economic status, family background, facial grade and dating and marriage decisions. Compared to the people in the census who married in the corresponding years, regular users of the matchmaking company are more likely to live in the metropolitan area (Seoul and neighboring cities) and are mostly college-educated. Conditional on education and region, their reported traits, such as income, height, and weight, are comparable to the general population. The empirical strategy we adopt is as follows. First, we evaluate all the photographs we

8

have obtained both from the matchmaking company (regular and temporary members) and from the plastic surgery consulting company (inquirers); we then construct our measure of facial attractiveness (henceforth called “beauty index”). By comparing the beauty index of the 56 inquirers before and after surgery, we compute the impact of plastic surgery on our measure of beauty (beauty index). Second, by comparing the beauty indexes of the 79 regular members with their facial grades, we estimate the relationship between the two beauty measures—our beauty index and the company’s facial grade. Using the estimated relationship between our beauty index and the company’s facial grade, we then calculate the effect of plastic surgery on facial grade. Third, using detailed information on more than 20,000 regular members of the matchmaking company, we estimate the extent to which facial attractiveness, as measured by facial grade, affects users’ dating and marriage outcomes as well as annual income. Finally, by combining the results from the second and the third steps, we compute the extent to which plastic surgery affects an individual’s outcomes in marriage and labor markets. Note that by studying the 61 temporary members who offered a self-reported facial grade, we attempt to control for “selfesteem” in estimating the beauty premium in the labor market. In addition, we conduct other robustness checks.

2.3 Beauty Index We measure facial attractiveness in a way similar to Biddle and Hamermesh (1998) and Mobius and Rosenblat (2006). Each photo is evaluated on a scale from 1 to 5, with 1 denoting the least attractive and 5 the most attractive. The reviewers of the photos are 50 senior undergraduate and graduate students from the Seoul National University, 50 percent of whom are female. On average, each reviewer evaluates about 117 randomly chosen photos and each photo is evaluated by 23 reviewers. We construct our measure of beauty Bj for individual j as follows: Let ri,j be the reviewer i’s rating of the photo of individual j; ri be the average of ratings for all the photos that the reviewer i rated; σi be the standard deviation of the ratings that the reviewer i rated; N and Nj be the total number of photos evaluated and the number of evaluators who evaluated photo j. We compute a standardized rating, Ri,j, by subtracting ri from the raw measure ri,j and then dividing it by σi. This procedure can account for the possibility that reviewers may have different perceptions of “average” beauty and of the difference in beauty associated with a unit difference in the measurement scale. Finally, we take the average of Ri,j over i and normalize its variance as

9

one, thereby obtaining our beauty index Bj.

B j = b j / ! b where b j =

1 Nj

!R

i, j

, Ri , j =

i

ri , j " ri

!i

, " b2 =

1 N

!b

2 j

(1)

j

This procedure allows us to interpret the regression coefficients on beauty index as the effect of a one-standard-deviation increase in physical attractiveness on the dependent variable. The top panel of Table 4 presents the summary statistics of the beauty index. The beauty index ranges from -2.4 to 2.6 with unit standard deviation.

2.4 Relationship between Beauty Index and Facial Grade The dataset that provides information about the beauty premiums in labor and marriage markets uses facial grade, a different measure of facial attractiveness that is assigned by the matchmaking company. We first quantify the relationship between our beauty index (Bj) and the facial grade (FGj) assigned by the matchmaking company using three statistical models: linear, ordered logit and ordered probit. After converting the company’s facial grades (A to D~F) to numeric values (5 to 2), we use them as continuous measures in the linear model while using them as categorical ones in the ordered models. FG*j = " + #1( j = female) + $B j + % j

(2)

where

!

for linear model,

FG j = FG*j

# FG j % % FG j for ordered models, $ % FG j % & FG j

= 2 if FG*j " c 2 = 3 if c 2 < FG*j " c 3 = 4 if c 3 < FG*j " c 4 = 5 if c 4 < FG*j

#' j ~ standard logistic dist. for ordered logit and $ &' j ~ standard normal dist. for ordered probit

As shown in columns (1) to (3) of the top panel of Table 5, our measure of beauty is positively correlated with the company’s facial grade, and the relationship is statistically significant. The ! bottom panel of the table presents the predicted distribution of the average individual’s facial grade. For example, according to the ordered probit model as reported in column (3), the average

10

man will be assigned facial grade D~F with probability 10.5 percent, C with probability 37.0 percent, B with probability 45.0 percent, and A with probability 7.6 percent. The prediction by the ordered logit model is similar to that of the ordered probit model. For the case using the OLS result, we convert the predicted continuous facial grade into the two nearest (discrete) facial grades and then compute the probability assigned to each grade with linear interpolation. The result from OLS is qualitatively similar to those from the ordered probit and logit in the sense that an individual with an average beauty index is most likely to get facial grade C among all facial grades. Our beauty measure is subject to measurement error, although we attempt to reduce the error by taking the average of multiple beauty ratings. To gauge the sensitivity of the estimated relationship between beauty index and facial grade, we employ alternative models that can address measurement errors in our beauty index. In particular, we assume that measurement errors in the beauty index are drawn from a normal distribution and are not correlated with errors governing the relationship between beauty index and facial grade (i.e., εj in Equation (2)). Then, we use maximum likelihood to re-estimate the relationship between beauty index and facial grade. The results are presented in columns (4) to (6) of Table 5. As expected, the models with measurement errors yield a larger estimated coefficient for beauty index than the corresponding one without measurement errors and predict a more concentrated distribution of predicted facial grade. However, the difference between the models with and without measurement errors is small, in terms of their prediction for the distribution of facial grades given a beauty index. For example, the coefficient of beauty index is higher with measurement error than without (0.413 vs. 0.358), and the probability of men with the average beauty index being assigned a facial grade of either B or C is slightly higher with measurement error than without (0.824 vs. 0.820).

3 Effect of Plastic Surgery on Beauty 3.1 Who Inquires About Plastic Surgery? We conduct two exercises to examine who demands plastic surgery: whether “pretty” wants to be prettier or “plain” wants to join pretty. First, we compare the average beauty index of inquirers with that of the average Korean. Inquirers are composed of the 56 individuals who have contacted the plastic surgery consulting company, whereas the average Koreans are the

11

synthesized photos in Figure 2 of Section 2. As shown in Table 4, the average beauty index of inquirers is -0.86 for men and -0.46 for women, whereas the index for the average Koreans is 1.79 for men and -0.37 for women. Second, we compare the 56 inquirers with the 140 users of the matchmaking company whose photos we have. The average beauty index is -0.25 for male users and 0.10 for female users. The average beauty index of male inquirers is statistically lower than that of the average man and also that of the male users at the conventional significance level. The average beauty index of female inquirers is not statistically different from that of the average woman but is lower than the average beauty index of female users at the conventional level. These results imply that, on average, the “plain” are more interested in plastic surgery— more so for the plain males.

3.2 Cost of Plastic Surgery Table 6 shows the average cost of plastic surgery given the quartile of an individual’s initial beauty index. We use two measures of cost. Cost 1 is based on the information provided by the plastic surgeons that made the dataset available to us; Cost 2 is based on an alternative website that provides the average costs of certain plastic surgery procedures in Korea.4 Since we know what procedures each individual inquires about, we can compute total surgery cost for each individual inquirer by summing up all the costs of the procedures asked about. The average total cost of plastic surgery procedures that individuals inquire about is 7,690,909 won based on the first measure and 6,294,444 won based on the second measure (roughly $7,690 and $6,294, respectively). We classify inquirers into four groups based on their initial beauty indexes and compare them in terms of the average cost of plastic surgeries. We do not find any statistically significant difference in the average surgery costs across groups.

3.3 Effect of Plastic Surgery on Beauty Table 6 presents the changes in the beauty index due to plastic surgery for those sample members who have contacted plastic surgeons. On average, plastic surgery increases the beauty index by 0.24 standard deviations for men and by 0.48 standard deviations for women. The extent of improvement resulting from plastic surgery diminishes, as the initial level of beauty

4

See http://www.sung-hyung.com. 12

gets higher. We consider the extent to which plastic surgery improves the relative beauty ranking by examining the individuals’ beauty quartiles before and after surgery (see rows titled “beauty index quartile after surgery” in Table 6). Out of 10 male inquirers at the lowest beauty index quartile (Q1), three move up to the second-lowest quartile (Q2) after plastic surgery, whereas the rest remain at the lowest quartile. On the other hand, women initially at the lowest quartile can possibly move to the second-highest quartile (Q3) with a 23.1 percent chance and the highest quartile (Q4) with a 7.7 percent chance. Table 7 examines the effect of plastic surgery using regression analysis. The dependant variable is the difference in the beauty index between aftersurgery and before-surgery, and independent variables include gender, initial beauty index, type of surgery, and associated cost. We find that plastic surgery statistically significantly increases women’s beauty indexes.5 Columns (2) to (4) show that the improvement of beauty due to plastic surgery decreases in the initial beauty level. Next, we convert the improvement in beauty index as a result of plastic surgery into improvement in terms of facial grade. This conversion is needed because we will use facial grade to measure the beauty premium in the labor and marriage markets in subsequent sections. As the improvement after surgery depends on the initial beauty index, we choose three levels of initial beauty index for both males and females (mean, 25th percentile and minimum of beauty index) and present the effect of surgery in terms of facial grade in Table 8. The table shows the beauty index before and after surgery in the top panel and associated changes in the distribution of facial grade using the results from the ordered probit regression. The table indicates that the improvement in beauty due to plastic surgery is positive for all three types of individuals but that the benefit decreases sharply in the initial beauty. For example, with plastic surgery, a man with the lowest beauty index can increase his chance to get facial grade B by 8.2 percentage points and facial grade A by 1.4 percentage points, whereas men with mean beauty can increase only by 0.8 percentage points for facial grade B and 0.4 percentage points for facial grade A. It is worth mentioning that the treatment effect of plastic surgery we discuss in this paper might be termed “treatment effect on potentially treated.” This is because we measure the treatment effect for those who are potentially interested in receiving the surgery. This concept lies between the “average treatment effect” and the “treatment effect on the treated,” in the sense 5

We find that the sum of the coefficient estimates for the constant term and female dummy variable in column (1) is statistically different from zero at the 1 percent level. 13

that the set of people who are potentially interested in receiving a treatment is larger than the set of people who actually receive the treatment but smaller than the entire set of people. Further discussions of measuring the effect of surgery on beauty are relegated to Section 6.

4 Beauty Premiums 4.1 The Beauty Premium in Marriage Market We examine the correlation between an individual’s facial grade and the individual’s spousal income among users who married through the online matchmaking company. In columns (1) and (3) of Table 9, we first control for typical explanatory variables (e.g., education, age, region) and regress the logarithm of spousal income on these variables and a user’s facial grade. Compared with an average user (facial grade C), we find that a male user with a facial grade A enjoys a spousal income premium of 18.8 percent, whereas an A-graded female user enjoys a spousal income premium of 12.7 percent. Next, we control for other characteristics that are rarely available in datasets used in other existing studies on beauty premiums: father’s educational attainment, parental wealth, and an education index. The education index, constructed by the matchmaking company, summarizes the ranking of a school and the major field of study based on the students’ average scores on the nation-wide college entrance examination (comparable to the SAT in the U.S.). Note that in Korea, students need to apply for a specific major field of study at the time of their college applications. We think that a user’s education index may capture future income prospects or socioeconomic status in addition to the user’s current income and level of educational attainment. Columns (2) and (4) show that even if we include additional control variables, an individual with a facial grade A or B enjoys significant spousal income relative to the average individual. When we convert the percentage premium into an absolute dollar amount, the annual spousal income of an A-graded man is 4,210,000 won (roughly $4,210) larger than that of an average man, whereas the spousal annual income of an A-graded woman is 5,770,000 won (roughly $5,770) larger than that of an average woman. However, we do not find any statistically significant difference in the spousal income between people with facial grade D~F and people with grade C. So far, we estimate the beauty premium in the marriage market conditional on an individual’s being married. If beautiful individuals have high reservation utility and get married

14

only if their spousal income is large enough, then our estimates of the beauty premium might face an upward bias due to selection. Addressing this selection issue is challenging because it requires an exogenous variation in the propensity of marriage that is not correlated with beauty. We tackle this issue indirectly as follows: Among users of the matchmaking company, a user with facial grade C is 5 percent more likely to get married than another user with facial grade A who is otherwise identical. Thus, we rank users with facial grade C in terms of spousal income and exclude those whose spousal income is less than the 5th percentile. We then re-estimate the beauty premium in terms of spousal income and find that beauty premium decreases, but not by much. For example, for an A-graded man, the beauty premium through the spousal income decreases from 18.8 percent to 15.6 percent, and for a B-graded man it decreases from 6.0 percent to 3.3 percent.

4.2 The Beauty Premium in Labor Market We regress the logarithm of the annual income of a user on his or her facial grade and other characteristics (see Table 10). Regression results show that both men and women whose facial grade is above the median (i.e., A or B vs. C) enjoy a beauty premium in terms of their income, while unattractive users (i.e., D to F) do not receive a monetary penalty as compared to the median-looking users. Columns (1) and (4) use control variables typical in the literature and find that an A-graded man and woman earn 15.2 percent and 11.1 percent higher incomes than their C-graded counterparts. Even after more control variables are included such as regional dummies, industries, education index and parental background, the magnitude of the beauty premium is still sizable, as reported in columns (2), (3), (5) and (6) of Table 10.

5 Returns to Plastic Surgery in Marriage and Labor Markets This section presents the expected monetary benefit of plastic surgery in marriage and labor markets and then examines whether individuals can recoup their plastic surgery cost and, if so, in how many years.

5.1 Expected Benefit of Plastic Surgery In the previous sections we examined the extent to which plastic surgery improves facial grades,

15

and the beauty premiums in marriage and labor markets. This section combines these results using the relation between the beauty and the facial grade measures, and it examines the returns to plastic surgery in marriage and labor markets. We consider three types of individuals in each gender based on their initial beauty level (mean beauty, 25th percentile, and 1st percentile) and compute the expected returns in terms of their own income and spousal income. For each type of individual, we predict his or her facial grade before the surgery and after the surgery using the results from Table 5. Then, using the estimated beauty premium for each value of the facial grade measure (results in Section 4), we calculate the expected monetary benefits of receiving the plastic surgery. The top panel of Table 11 shows the results. Consider the rows titled “Benefit.” A man in the 1st percentile of the beauty index can increase his own income by 0.79 percent and spousal income by 0.68 percent. For all outcome variables, the expected benefits decrease in the initial beauty index.

5.2 Cost-Benefit Analysis In this section, we compute the expected lifetime monetary benefit from plastic surgery for three types of individuals in terms of their initial beauty indexes (mean, 25th percentile, and minimum beauty index). We consider those individuals who are college educated and have average sample characteristics except for age, beauty, and education. For the purpose of computing a discounted lifetime benefit from plastic surgery, we first choose a set of generous, but still plausible, assumptions: Each individual receives plastic surgery just before starting his or her career. An individual starts to work at age 25 for men and age 23 for women and keeps working until age 65.6 We assume that college graduates get married at the age of 31 for men and 28 for women, which are the average ages of the first marriages in Korea in 2008. To aggregate the benefit stream over the life-cycle, we use a discount factor of 1 (i.e., zero discount rate). Finally, we assume that plastic surgery costs 6,036,840 won for men and 6,434,290 won for women, based on the lower cost measure. Using the Mincerian wage regression results in columns (1) and (4) of Table 10, we construct the expected life-cycle wage profile of college graduates for each of the three beauty index levels. Similarly, we compute the life-cycle wage profile of the college graduates’ spouses using the results from a spousal income regression in columns (1) and (3) of Table 9. 6

The two-year gap is due to men’s mandatory military service in Korea. The legal retirement age in Korea is 65. 16

Then, at each age of an individual, we compute the net benefit up to the age by subtracting the surgery cost from the cumulative monetary benefit up to that age. This age-benefit profile allows us to compute how long it would take for an individual to recoup the cost of plastic surgery given his or her initial beauty. For example, if an individual’s net benefit at age 65 is negative, then we will conclude that he or she cannot recoup the cost before his or her retirement. If the net benefit turns to be positive at a certain age, we compute the years required to recoup the cost by linearly interpolating the last age of the negative net benefit and the first age of positive net benefit. The bottom panel of Table 11 shows the results. If we use the ordered probit model to convert our beauty index to facial grade (row titled “Probit”), men with the minimum value of beauty index can recoup the cost of surgery in 13 years (i.e., by age 38), and the 25th percentile men in 20 years, whereas men with mean beauty cannot recoup the cost before their retirement at age 65. Similar results are shown for women. These results remain similar if we use the ordered logit model or OLS to convert our beauty index to facial grade. Our finding of low monetary returns for plastic surgery is anticipated from the earlier findings because plastic surgery mostly helps unattractive people obtain average beauty and that there is little beauty premium from having average, rather than low, beauty. Our finding is also comparable to the findings in Hamermesh et al. (2002). They study women’s spending on cosmetics and clothing in Shanghai, China, and find that additional spending on those goods increases a woman’s perceived beauty but generates little monetary benefits in the form of higher earnings. Because we find that the monetary benefit from plastic surgery is generally small relative to the cost, we compute how long it would take to break even if a person enjoys the maximum benefit from the plastic surgery (from facial grade D~F to grade A). The results show that it will take 1.3 years for men and 2.5 years for women (columns (4) and (8) of Table 11). As to the reason why it takes longer for women than men: first, women do not enjoy spousal income for the initial 5 to 6 years before marriage and, secondly, because the returns to a facial grade A are lower for women than for men.

17

6. Robustness Checks and Discussions 6.1 Use of Hypothetical Looks Our measure of the extent to which plastic surgery improves facial attractiveness is based on the comparison of the photo images before and after plastic surgery procedures, where the before image is an actual photo but the after look is a hypothetical one. Using hypothetical looks is not necessarily more problematic compared to using an actual photo of people who got plastic surgery. It is reasonable to expect that an individual who expects a large benefit from plastic surgery will be more likely to actually get the surgery because many plastic surgeons in Korea have software programs to generate hypothetical looks. Thus, we may be more vulnerable to selection bias if we measure the effect of plastic surgery on beauty only based on the actual photos. Although it is not perfect, using hypothetical looks may relieve the selection bias problem. Still, there remains a valid concern. Hypothetical looks may overstate the true effect of plastic surgery procedures as plastic surgeons may have a tendency to exaggerate the benefits of plastic surgery. The best way to check for this problem is to compare actual after-surgery photos with the corresponding hypothetical ones for those who have self-selected themselves to go through the surgery procedures. However, this option is not feasible because in Korea plastic surgeons are not allowed to share their patients’ identities (including full head-to-shoulder photos) with a third party. As an alternative to this best option, we use photos of Korean celebrities who publicly state that they have undergone plastic surgery and make available online their actual photos before and after the plastic surgery. The measured effect of plastic surgery using the actual before and after photos of celebrities may give us an upper bound of the true effect because if the surgery had not been successful, then they might not have become celebrities after the surgery or might have not even stayed in the entertainment business. We choose five male and five female celebrities whose photo qualities in terms of resolution and size are comparable to the photos in our datasets.7 We evaluate their facial beauty before and after surgery and report the effect of plastic surgery on celebrities in the bottom panel of Table 6. Conditional on their initial facial beauty, the improvement due to plastic surgery for celebrities 7

The reason we evaluate only 10 celebrities is because until recently most celebrities in Korea have denied any allegation of having plastic surgery, resulting in only small number of celebrities whose photos before and after surgery are available. 18

exceeds that for non-celebrities, suggesting that the hypothetical looks after plastic surgery do not seem to greatly overstate the benefit of surgery.

6.2 Alternative Assumptions for Cost-Benefit Analysis In this section, we change our assumptions used in the cost-benefit analysis and examine the sensitivity of our findings. First, we redo our cost-benefit analysis using the estimation results from measurement error models. As shown in the row titled “Probit, measurement error” in Table 11, the result for the cost-benefit analysis qualitatively remains the same. Second, we use beauty premiums in marriage and labor markets after controlling for family background and a more detailed education index, shown in columns (2) and (4) of Table 9 and columns (3) and (6) of Table 10. We then redo the cost-benefit analysis, keeping the rest of the assumptions the same as those described in Section 5. Using the ordered probit model, we find that using the alternative beauty premium increases the number of year to recoup the cost of surgery by from half a year to 7 years (row titled “Probit, family background” in Table 11). Third, we change several assumptions: We change the discount factor from 1 to an alternative value of 0.95 based on the average interest rate of a one-year Korean Treasury bond from 2000 to 2008; we use the beauty premiums that are estimated after controlling for family background; and we assume that people can enjoy only 50 percent of spousal income in that one dollar in one’s spouse’s pocket is only worth as much as 50 cents in one’s own pocket. Under these assumptions, we find that an individual cannot recoup his or her expense unless his or her facial grade changes from D~F to A (see the last row of Table 11).

6.3 Beauty and Self-Esteem In Section 4, we estimate beauty premiums by controlling for individuals’ characteristics that are rarely available in datasets used in the literature (parental educational attainment and parental wealth). Therefore, our estimates may be less vulnerable to a concern that the beauty premiums may mainly reflect other unobservable traits that are correlated with beauty. However, one may still be concerned about the possibility that the estimated beauty premium may largely reflect self-esteem or other unobservable traits. Although it is impossible to directly address this concern, we examine the extent to which self-esteem may account for the estimated beauty premium in labor market as follows. In our dataset, we have 61 temporary users of the company 19

who self-reported their facial grades. We have an objective measure of their beauty constructed from other people’s evaluations. If people with high self-esteem are more likely to exaggerate their facial grade relative to those with low self-esteem, then the difference between our measure of beauty and the self-reported beauty will be positively correlated with self-esteem. Following this idea, we construct a variable that is a ratio of the self-reported grade over our beauty index. This variable is positively correlated with the extent to which a user considers his or her beauty relative to a third-party’s evaluations of the user’s facial features. We regress a user’s logarithm of income on his/her beauty index, gender, age, age squared and educational attainment with and without additionally controlling for our measure of self-esteem. If most beauty premium in the labor market arises from self-esteem, we may find the coefficient of our beauty index with controlling for the self-esteem measure will be much smaller than that from the regression without. We find that the estimated coefficient of our beauty index remains the same regardless of whether we control for self-esteem (3.673 vs. 3.637).

7. Conclusion This study quantifies the effect of plastic surgery on beauty, and it examines the monetary benefit in marriage and labor markets that results from beauty improvement relative to the cost of plastic surgery. Our results on the potential benefits of plastic surgery must be interpreted with caution. First, our cost-benefit analysis does not account for either the non-monetary benefits of plastic surgery or the risk of having unsuccessful surgery. For example, we find that an individual with attractive facial features is more likely to be asked out by the opposite sex. It is possible that the utility from the increased popularity in the dating/marriage market may outweigh the disutility from paying surgery cost, in particular for those whose initial level of beauty is low. Secondly, our analysis of beauty premiums may not hold true if there is a large change in the distribution of beauty in the society. For example, if most people get plastic surgery, then beauty premiums may cease to exist in marriage and the labor markets.

References American Society of Plastic Surgeons. 2008. “2007 Statistics National Clearinghouse of Plastic Surgery Statistics.” Available at http://www.plasticsurgery.org/ 20

Banerjee, Abhijit V., Esther Duflo, Maitreesh Ghatak and Jeanne Lafortune. 2009. “Marry for What? Caste and Mate Selection in Modern India.” MIT Department of Economics Working Paper No. 09-14. Bergstrom, Theodore C. and Mark Bagnoli. 1993. “Courtship as a Waiting Game.” The Journal of Political Economy, 101(1): 185-202. Biddle, Jeff E. and Daniel S. Hamermesh. 1998. “Beauty, Productivity, and Discrimination: Lawyers’ Looks and Lucre.” Journal of Labor Economics, 16(1): 172-201. Case, Anne and Christina Paxon. 2008. “Stature and Status: Height, Ability and Labor Market Outcomes.” Journal of Political Economy, 116(3): 499-532. Chiappori, Pierre-Andre, Murat F. Iyigun and Yoram Weiss. Forthcoming. “Investment in Schooling and the Marriage Market.” American Economic Review. Deaton, Angus S. and Raksha Arora. 2009. “Life at the Top: The Benefits of Height.” NBER Working Paper 15090. Fackler, Martin. 2009. “Economy Blunts Korea’s Appetite for Plastic Surgery.” New York Times Magazine. January 1, available at http://www.nytimes.com/2009/01/02/business/worldbusiness/02plastic.html Fisman, Raymond, Sheena S. Iyengar, Emir Kamenica and Itamar Simonson. 2006. “Gender Differences in Mate Selections: Evidence from a Speed Dating Experiment." The Quarterly Journal of Economics, 121(2): 673-697. Glied, Shery and Matthew Neidell. 2008. “The Economic Value of Teeth.” NBER Working Paper 13879. Hamermesh, Daniel S. and Jeff E. Biddle. 1994. “Beauty and the Labor Market.” American Economic Review, 84(5): 1174-1194. Hamermesh, Daniel S., Xin Meng and Junsen Zhang. 2002. “Dress for Success—Does Primping Pay?” Labor Economics, 9(3): 361-373. Hitsch, Guenter J., Ali Hortacsu and Dan Ariely. 2009. “What Makes You Click? Mate Preferences and Matching Outcomes in Online Dating.” MIT Sloan Research Paper 4603-06. Hitsch, Guenter J., Ali Hortacsu and Dan Ariely. Forthcoming. “Matching and Sorting in Online Dating Markets.” American Economic Review. International Society of Aesthetic Plastic Surgery. 2004. “ISAPS Statistics 2004.” Available at http://www37.tok2.com/home/koreanworld/data/archives/surgery/stats2004res.html

21

Lafortune, Jeanne. 2009. “Making Yourself Attractive: Pre-Marital Investments and the Returns to Education in the Marriage Market.” Working paper, University of Maryland, College Park. Lee, Soohyung. 2009. “Marriage and Online Match Search Services.” Working paper, University of Maryland, College Park. Mobius, Markus M. and Tanya S. Rosenblat. 2006. “Why Beauty Matters.” American Economic Review, 96(1): 222-235. Persico, Nicola, Andrew Postlewaite and Dan Silverman. 2004. “The Effect of Adolescent Experience on Labor Market Outcomes: The Case of Height.” Journal of Political Economy, 112(5): 1019-1053. Pfann, Gerard A., Jeff E. Biddle, Daniel S. Hamermesh and Ciska M. Bosman. 2000. “Business Success and Business Beauty Capital.” Economic Letters, 67(2): 201-207. Rhee, Seung Chul. 2006. “The Average Korean Attractive Face.” Aesthetic Plastic Surgery, 30(6): 729-730. Rhee, Seung Chul, Eun Sang Dhong and Eun Sik Yoon. 2009. “Photogrammetric Facial Analysis of Attractive Korean Entertainers.” Aesthetic Plastic Surgery, 33(2): 167-174. Rhee, Seung Chul and Sang Hwan Koo. 2007. “An Objective System for Measuring Facial Attractiveness.” Plastic and Reconstructive Surgery, 119(6): 1952-3 and author reply 1953-4. Photos for before and after plastic surgery available at http://iaanclinic.com/welcome.htm Wyer, E. Bingo. 1998. The Unofficial Guide to Cosmetic Surgery. New York: John Wiley & Sons. Documents Written in Korean Cho, Yong Jin. 2007. Facial Features of Koreans. Seoul: Hainaim. Available at: http://www.kyobobook.co.kr/product/detailViewKor.laf?ejkGb=KOR&mallGb=KOR&barcode= 9788973378593&orderClick=LEA ELLE Korea. 2007. “Survey of Lifestyle of Korean Women.” November 6. Available at http://www.elle.co.kr/life/lifeStyleView.html?AI_IDX=2330 Korea Social Research Center. 2004. “Survey of Students at Primary and Secondary Schools in Korea.” Press Release. Available at http://ksrc.or.kr/ Lim, In Sook. 2002. “The Body Project in Korea: Focusing on Cosmetic Surgery Industry.” Journal of Korean Sociology, 36(3): 138-204. Available at http://www.ksa.re.kr/bbs/skin/sirini_ezset_fullpack/ezset_catch_trackback.php?id=zine&no=10

22

National Statistics Office. 2009. “Trend of Marriage in 2008.” Press Release. Available at http://kostat.go.kr/board_notice/BoardAction.do?method=view&board_id=144&seq=60&num=6 0&catgrp=nso2009&catid1=k04___0000&catid2=k04b__0000&catid3=k04ba_0000&catid4= Ministry of Health, Welfare and Family Affairs. 2004. Press Release. Available at http://www.mw.go.kr/front/al/sal0101vw.jsp?PAR_MENU_ID=04&MENU_ID=040101&BOA RD_ID=110&BOARD_FLAG=00&CONT_SEQ=28788&page=1 Uhm, Hyun Shin. 2008. “Perception on Beauty and Plastic Surgery.” PhD dissertation. Kyunghee University.

23

Table 1: Use of Plastic Surgery Survey

Sample

Ministry of Health, Welfare and Family Affairs (2004) Korea Social Research Center (2004) Uhm (2007)

2,034 college students (1,565 female) 3,861 students in primary or secondary schools 810 women over 18 living in Seoul and Kyunggi 9,324 women aged 20 to 35 living in Seoul -

ELLE Korea (2007) ARA Consulting (2009)

Fraction of people who had plastic surgery 52.5% of women 5.8% of men 37.4% of women 47.1% 76.1% 30% of women aged 20 to 50

Table 2: Popular Plastic Surgery Procedures

Country Sample The five most popular procedures

(1) South Korea 2,034 college students Asian eyelid surgery (25.3%) Removing moles (22.0%) Nose augmentation (3.6%) Chemical peel (2.5%) Laser hair removal (2.5%)

(2) U.S. All procedures conducted in 2008 Botox (39.0%) Hyaluronic acid (8.9%) Chemical peel (8.7%) Laser hair removal (7.7%) Microdermabrasio n (7.6%)

(3) U.S. Patients aged 20 to 29 in 2008 Laser hair removal (26.2%) Breast augmentation (14.0%) Nose reshaping (11.0%) Botox (10.0%) Microdermabrasion (9.4%)

Source: Ministry of Health, Welfare and Family Affairs (2004) for column (1); American Society of Plastic Surgeons (2008) for columns (2) and (3).

24

Table 3: Summary Statistics of Datasets Data source

Number of individuals Number of photos Fraction of men Fraction of regular users Average age Membership status Facial grade [No. obs.] - D~F -C -B -A

Plastic surgeon

(1) 56 112 35.71%

Matchmaking company Matchmaking company Users in mid Users between 2002 -August 2008 and June 2006 (2) (3) 140 20,689 140 0 57.14% 46.10% 56.43% 100.00% 30.74 31.36 Regular Online Regular Assigned Self-reported Assigned [79] [61] [20,689] 8.86% 22.95% 11.89% 32.91% 36.07% 42.71% 46.84% 36.07% 38.30% 11.39% 4.92% 7.10%

Table 4: Distribution of Beauty Index

Sample - All - Men - Women Non-inquirers - All - Men - Women Surgery inquirers (before) - All - Men - Women Surgery inquirers (after) - All - Men - Women

No. obs. (1)

Mean (2)

S.D. (3)

Min. (4)

Max. (5)

252 120 132

-0.240 -0.417 -0.078

1.000 1.007 0.969

-2.481 -2.481 -2.383

2.662 2.662 1.932

140 80 60

-0.103 -0.253 0.097

0.927 0.919 0.907

-2.431 -2.431 -1.943

1.932 1.884 1.932

56 20 36

-0.607 -0.863 -0.464

1.089 1.137 1.051

-2.285 -2.155 -2.285

2.662 2.662 1.740

56 20 36

-0.215 -0.628 0.015

1.010 1.085 0.900

-2.481 -2.481 -2.383

2.453 2.453 1.891

25

Table 5: Relationship between Beauty Index and Facial Grade Dependant variable: Facial grade

No measurement error OLS Ordered Ordered logit probit (1) (2) (3) Regression Beauty index (Bj) Female Constant

0.249** (0.095) 0.005 (0.180) 3.604*** (0.132)

Cutoffs c2 (D~F vs. C) c3 (C vs. B) c4 (B vs. A) σ (rating) No. obs. R2/ log likelihood Prediction Average men Pr(D~F) Pr(C) Pr(B) Pr(A) Average women Pr(D~F) Pr(C) Pr(B) Pr(A)

79 0.085

0.634** (0.247) -0.036 (0.435)

0.358*** (0.254) 0.017 (0.250)

-2.473*** -1.404*** (0.478) (0.247) -0.381 -0.213 (0.335) (0.198) 2.153*** 1.287*** (0.438) (0.239) 79 79 -90 -90

Measurement error Ordered Ordered logit probit (4) (5) (6)

OLS

0.284*** (0.105) -0.015 (0.178) 3.615*** (0.130)

0.733*** (0.286) -0.082 (0.442)

0.413*** (0.158) -0.011 (0.254)

2.049 79 -3,529

-2.525*** (0.488) -0.412 (0.341) 2.152*** (0.442) 2.049 79 -3,527

-1.433*** (0.253) -0.231 (0.201) 1.284*** (0.241) 2.049 79 -3,527

0.000 0.500 0.500 0.000

0.099 0.372 0.447 0.082

0.105 0.370 0.450 0.076

0.000 0.503 0.497 0.000

0.098 0.375 0.448 0.079

0.104 0.373 0.451 0.073

0.000 0.415 0.585 0.000

0.084 0.343 0.477 0.096

0.082 0.338 0.483 0.097

0.000 0.407 0.593 0.000

0.084 0.348 0.476 0.092

0.082 0.343 0.482 0.092

Standard errors are in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%.

26

Table 6: Cost and Effect of Plastic Surgery Initial beauty Avg. Q1 Q2 No. of obs. - Men - Women Cost a - Men Cost 1 Cost 2 - Women Cost 1 Cost 2 Beauty index - Men Before surgery Changes after surgery Beauty index quartile after surgery Q1 Q2 Q3 Q4 - Women Before surgery Changes after surgery Beauty index quartile after surgery Q1 Q2 Q3 Q4 Celebrities (Changes after surgery) - Men - Women

Q3

Q4

20 36

10 13

6 9

3 9

1 5

775 (479) 604 (398)

800 (432) 620 (406)

722 (447) 583 (346)

433 (76) 320 (69)

1900 (0) 1430 (0)

766 (515) 643 (401)

733 (698) 643 (456)

607 (381) 513 (296)

950 (566) 763 (443)

777 (490) 651 (432)

-0.863 (1.137) 0.235 (0.461)

-1.654 (0.479) 0.275 (0.364)

-0.654 (0.135) 0.371 (0.421)

0.177 (0.327) -0.022 (0.847)

2.662 (0.000) -0.209 (0.000)

-

7 3 0 0

0 3 3 0

0 1 1 1

0 0 0 1

-0.464 (1.051) 0.479 (0.861)

-1.580 (0.461) 0.938 (0.740)

-0.427 (0.257) 0.647 (0.568)

0.151 (0.253) 0.180 (0.872)

1.264 (0.370) -0.479 (0.763)

-

5 4 3 1

1 2 3 3

0 4 1 4

0 0 2 3

2.382 (0.389)

0.238 (0.591) 1.074 (0.000)

0.238 (0.591) 2.612 (1.315)

3.610 (1.518)

Standard deviations are in brackets. Q1 stands for individuals who belong to the lowest quartile of beauty index (1st to 25th percentile); Q2 for 26th to 50th percentile; Q3 for 51st to 75th percentile; and Q4 for 76th to 100th percentile. aThe unit of cost is 10,000 Korean won, roughly 10 U.S. dollars.

27

Table 7: Effect of Plastic Surgery on Beauty Index Dependant variable: Change in beauty index after plastic surgery

Constant Female BBj

(1) 0.235 (0.167) 0.244 (0.208)

Log cost Nose Eye Face reshapinga No. obs. b) R2/log likelihood σ(rating)

56 0.025 -

No measurement errors Measurement errors (2) (3) (4) (5) (6) -0.055 -0.686 -0.245 -0.107 -0.306 (0.163) (0.931) (0.248) (0.163) (1.068) 0.378** 0.394** 0.375** 0.552*** 0.555*** (0.186) (0.193) (0.187) (0.183) (0.184) -0.336*** -0.347*** -0.367*** -0.380*** -0.379*** (0.083) (0.085) (0.088) (0.126) (0.126) 0.095 0.030 (0.139) (0.162) 0.169 (0.191) -0.057 (0.196) 0.281 (0.191) 56 56 55 1,304 1,281 0.257 0.264 0.296 -57,132 -56,643 2.166 2.162

a

Face reshaping procedures change the facial shape into an oval one. The unit of observation is a photo in columns (1) to (4) and a (photo, reviewer) combination in columns (5) and (6). Standard errors are in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. b

Table 8: Effect of Plastic Surgery on Facial Grade Men Mean (1) Beauty index - Before - After c Ordered probit - After – Before Pr(FG = D~F) Pr(FG = C) Pr(FG = B) Pr(FG = A)

25th a) (2)

Women 1st b) (3)

Mean (4)

25th a) (5)

1st b) (6)

-0.417 -1.083 -2.481 -0.078 -0.759 -0.332 -0.774 -1.703 0.271 -0.181

-2.351 -1.238

-0.005 -0.025 -0.090 -0.017 -0.038 -0.007 -0.019 -0.007 -0.031 -0.045 0.008 0.032 0.082 0.025 0.053 0.004 0.012 0.014 0.023 0.029

-0.117 -0.028 0.119 0.026

a

25th percentile. b st 1 percentile. c Results in column (2) of Table 7 are used.

28

Table 9: Beauty Premium in the Marriage Market Log spousal income a First date b Men Men Women Women Men Women (1) (2) (3) (4) (5) (6) 1(FG = A) 0.188*** 0.162*** 0.127*** 0.091* 0.044*** 0.118*** (0.062) (0.062) (0.048) (0.048) (0.012) (0.010) 1(FG = B) 0.060** 0.043 0.094*** 0.081*** 0.038*** 0.054*** (0.029) (0.029) (0.028) (0.028) (0.006) (0.006) 1(FG = D~F) 0.009 -0.003 0.021 0.023 -0.034*** -0.029*** (0.042) (0.041) (0.045) (0.046) (0.008) (0.008) 1(High school or less) -0.018 0.1 -0.099** 0.025 0.039** 0.016 (0.085) (0.085) (0.047) (0.062) (0.017) (0.013) 1(Tech. college) -0.082 0.218*** -0.077** 0.06 0.076*** 0.004 (0.058) (0.083) (0.036) (0.060) (0.014) (0.012) 1(Master’s or Ph.D.) 0.024 -0.046 0.099** 0.069 0.003 -0.024*** (0.032) (0.035) (0.043) (0.043) (0.007) (0.008) Education index 0.060*** 0.024** 0.022*** -0.006*** (0.012) (0.011) (0.002) (0.002) Log income -0.007 0.011 0.043*** -0.011 (0.049) (0.047) (0.009) (0.009) 1(Dad = college+) -0.016 0.083*** 0.022*** -0.003 (0.029) (0.029) (0.006) (0.005) Log parental wealth 0.001 -0.002 -0.023*** -0.028*** (0.008) (0.009) (0.002) (0.002) Age (10 yr.) 0.717* 0.61 0.207 0.27 -0.431*** -0.417*** (0.406) (0.398) (0.488) (0.484) (0.080) (0.084) Age squared -0.07 -0.055 0.017 0.008 0.056*** 0.061*** (0.056) (0.055) (0.077) (0.076) (0.011) (0.013) No. obs. 469 469 469 469 5,820 6,959 R2 0.093 0.162 0.159 0.193 0.119 0.136 a

The unit of annual income is 10,000 won (roughly equivalent to 10 U.S. dollars at the time). Fraction of partners accepting a first date. Columns (2), (4), (5) and (6) include region dummies. Standard errors are in parentheses. *significant at 10%; **significant at 5%; ***significant at 1%.

b

29

Table 10: Beauty Premium in the Labor Market Dependant variable: Log of annual income

1(FG = A) 1(FG = B) 1(FG = D~F) 1(High school) 1(Tech. college) 1(Master’s or Ph.D)

(1) 0.152*** (0.017) 0.062*** (0.009) -0.008 (0.011) 0.023 (0.022) -0.099*** (0.012) 0.068*** (0.010)

Men (2) 0.148*** (0.017) 0.056*** (0.009) -0.012 (0.011) 0.022 (0.022) -0.098*** (0.012) 0.066*** (0.010)

0.859*** (0.113) -0.070*** (0.016) 5,992 0.197

0.841*** (0.113) -0.068*** (0.016) 5,992 0.201

Education index 1(Dad = college+) Log parental wealth Age (10 yr.) Age squared No. obs. R2

(3) 0.096*** (0.016) 0.031*** (0.008) -0.01 (0.011) 0.145*** (0.023) 0.117*** (0.019) 0.021** (0.010) 0.042*** (0.003) -0.005 (0.008) 0.018*** (0.002) 0.880*** (0.108) -0.074*** (0.015) 5,990 0.297

(4) 0.111*** (0.014) 0.059*** (0.008) -0.002 (0.012) -0.153*** (0.014) -0.136*** (0.010) 0.058*** (0.011)

Women (5) 0.096*** (0.014) 0.042*** (0.008) -0.006 (0.012) -0.148*** (0.014) -0.133*** (0.010) 0.053*** (0.010)

0.561*** (0.122) -0.053*** (0.019) 7,160 0.107

0.527*** (0.120) -0.050*** (0.019) 7,160 0.134

(6) 0.084*** (0.014) 0.036*** (0.008) -0.01 (0.011) 0.023 (0.018) 0.078*** (0.017) 0.008 (0.010) 0.053*** (0.003) -0.021*** (0.007) 0.011*** (0.002) 0.516*** (0.114) -0.043** (0.018) 7,160 0.230

All columns except for (1) and (4) include regional dummies. Columns (3) and (6) additionally include industry dummies. Standard errors are in parentheses. *significant at 10%; **significant at 5%; ***significant at 1%. Note that the unit of annual income is 10,000 won (roughly equivalent to 10 U.S. dollars at the time).

30

Table 11: Cost-Benefit Analysis for Plastic Surgery

Benefit c Probit d - log own income - log spousal income - Pr(accepted for a 1st date) Years to recoup the cost - Probit d) - Logit d) - OLS d) - Probit, measurement error e - Probit, family background e - Probit, ρ = 0.95, family background, 50% of spousal income e

1st b (3)

D→A (4)

Mean (5)

Women 25th a 1st b (6) (7)

D→A (8)

0.399 0.393 0.265

0.794 0.677 0.753

16.000 17.900 9.400

0.408 0.493 0.334

0.646 0.791 0.554

1.015 1.205 1.069

11.300 10.600 12.400

20.361 19.717 18.249 18.943 29.470 -

12.922 12.685 10.423 11.741 18.273 -

1.281 1.281 1.281 1.281 1.958 2.010

26.633 25.872 22.847 19.310 33.023 -

19.191 18.502 15.957 14.733 23.793 -

14.013 13.819 10.168 11.518 17.062 -

2.494 2.494 2.494 2.494 2.994 3.163

Mean (1)

25th a (2)

0.119 0.124 0.067 -f -

Men

a

25th percentile. 1st percentile. c The statistics are presented multiplied by 100. For example, for mean beauty men, the probability of being accepted by the other party for a first date increases by 0.067 percentage points after the plastic surgery. d See Section 5.2 for assumptions used in the simulation. e See Section 6.2 for assumptions used in the simulation. f “-” refers to the case that an individual cannot recoup the cost of plastic surgery by the retirement age of 65. b

31

Appendix A. Description of Dataset from the Online Matchmaking Company The dataset is from a major Korean online matchmaking company, which helps its users to find a spouse among other users of the opposite sex. It has 20,689 individuals who started to use the company's services between January 2002 and June 2006 including their individual characteristics, stated marital preferences and history of dating outcomes. The annual membership costs 900,000 won in 2007 (approximately $900), which is about 3.5 percent of the average annual income in Korea. Because of the high membership cost, we think it is reasonable to assume users are primarily motivated to seek marriage rather than casual dating. The information about user characteristics is subject to several steps of verification by the company. As much as possible, key information is legally verified (e.g., age, education, employment, marital status) or independently evaluated by the matchmaking company (e.g., a facial grade). Self-reported user attributes, such as income and height, are monitored via user feedback. The company routinely surveys its users about their experiences and asks them to verify the reliability of other users' information. The matchmaking company's contract specifies that the service will be terminated if a user is found to provide incorrect information. We use four separate nationally representative datasets because no single populationbased dataset captures all the features observed in our data. The closest analog to the matchmaking dataset is the marriage register (MR). The MR is the population of newlyweds in South Korea in a given year and provides information about husband and wife's age, education, residence, hometown and marital history (never married vs. not). The three other datasets that we use are: the Basic Statistics Survey of Wage Structure (WS) for industries and income, the National Household Income and Expenditure Survey (HIS) for income of husbands and wives and the Survey of Physical Traits of Koreans (PT) for height and weight. We find that, in terms of observable traits, the users of the company represent a wide spectrum of Koreans, although the distribution of traits among users and the population is not the same. As shown in Tables A.1 and A.2, the users include all types of Koreans, in terms of marital status, educational attainment, geographical location and industry. The users, however, overrepresent those who are older, more educated, and currently live in, or are originally from, the capital or its surroundings (Seoul and Gyeonggi). The middle and bottom panels of Table A.2

32

compare self-reported income and physical traits of the users with those of the general population. For income comparison, we apply weights on the WS so that the weighted distribution of WS peoples’ characteristics is comparable to that of the users in terms of age, gender and educational attainment. The average annual income in the population whose characteristics are adjusted to be the same as users is 30 million won (about $30,000), whereas the average income among users is over 40 million won. Once we exclude people whose reported income is less than 5th percentile or more than 95th percentile among users, the trimmed mean income is comparable to that in the population. The average height and weight of the matchmaking company's users are remarkably similar to those in the PT.

33

Table A.1: Users’ Characteristics I This table compares characteristics of users in the matchmaking data set with the official marriage register (MR).

Year Number of individuals Composition (percentage) Women Divorced Non-Korean Age 26 and younger 27-29 30-33 34 and older Educational attainment Middle school or less High school College or more Technical college University Master’s and Ph.D. Region Seoul or Gyeonggi Gangwon Chungcheong Jeolla Gyeongsang Jeju and others Hometown Seoul or Gyeonggi Gangwon Chungcheong Jeolla Gyeongsang Jeju and others

Matchmaking dataset January 2002 ~ June 2006 All Married 20,689 1,594

MR 2002~2005 2,477,648

53.90 10.70 0.00

50.00 12.57 0.00

50.00 18.82 4.87

9.01 25.28 40.05 25.66

5.83 24.76 43.61 25.8

28.79 28.08 21.84 21.31

0.87 6.63 92.50 13.65 61.25 17.60

0.09 8.06 91.86 12.70 64.83 14.33

5.14 38.27 56.59 -

75.92 0.55 4.44 3.34 11.39 4.35

77.65 0.57 5.00 3.46 13.25 0.06

51.44 2.79 9.59 9.63 25.15 1.40

45.12 3.26 10.65 13.60 25.86 1.51

42.48 3.79 11.76 14.58 26.11 1.29

27.36 4.86 15.47 19.32 31.61 1.38

34

Table A.2: Users’ Characteristics II This table compares users of the matchmaking service with the general population. For population data, the top panel uses the WS (2002-2006) and the bottom panel uses the PT (2004).

Jan. 2002~June 2006

General Population WS (2002-2006)

0.04 20.37 9.23 4.26 4.74

7.92 16.36 6.27 10.54 19.32

9.41 10.19 0.76 20.32 9.55 5.6 2.41 3.12

5.49 5.17 12.69 11.01 3.02 2.2 -

4,054.63 3,138.08 3,137.05

3,046.49 N.A N.A.

Matchmaking dataset Year Distribution across industries (percentage) Agriculture, forestry, fishing, mining Manufacturing Public, electric power, gas, water supply Construction Wholesales and retail trade, consumer goods, restaurants and hotels Transportation, storage, communication Finance and insurance Real estate rental and business services Education services Health and social welfare Entertainment, housekeeping, personal service International and other foreign institution Others or unemployed Annual income (10,000 won) Mean Mean between 5th and 95th percentiles Median Gender-specific physical traits Height (feet, inches) 34 and younger Men Women 35 and older Men Women Weight (lbs.) 34 and younger Men Women 35 and older Men Women Body Mass Indexa 34 and younger Men Women Men 35 and older Women

PT (2004) 5’ 9” 5’ 4” 5’ 8” 5’ 4”

5’ 8” 5’ 3” [5’ 4”, 5’ 7”] b 5’ 2”

153.7 111.4 153.2 112.0

[153.2 , 157.0] [116.0 , 120.4] [151.9 , 158.3] [123.9 , 131.0]

22.8 19.0 23.0 19.4

[22.6 , 24.0] [20.3 , 21.7] [24.7 , 25.0] [22.8 , 25.1]

a. B.M.I. = 703 * weight (pounds) / (height (inches))2 b. [c,d] denotes the case where the corresponding statistic ranges from c to d.

35