Are Assets n Medcal Savngs Accounts Dscounted? ------ Evdence from a Natural Experment n Chna Maoyong Fan, Zhen Le and Guoen Lu November 23rd, 2010 Abstract: In Chna, Medcal Savngs Accounts (MSAs) are a major tool fnancng health care consumpton n urban areas. Whether MSAs control medcal expendtures and encourage savng s based on an assumpton that enrollees treat the MSA money the same as ther pocket money. Ths assumpton has never been tested. Gven the mandatory and restrctve nature of MSAs n Chna, we hypothesze that enrollees may dscount ther MSAs and spend them prematurely. To test whether assets n MSAs are dscounted, we take advantage of a polcy change as a natural experment n cty of Zhenjang. The polcy change affected dfferent age cohorts dfferently n terms of fnancal contrbutons to MSAs. Emprcal results show that a reducton n MSAs caused enrollees to reduce ther annual medcal expendtures by more than the amount of the MSA reducton. The effect was largest for those wth ntermedate medcal expendtures, who were more lkely to exhaust ther MSAs and pay out-of-pocket expenses. The results are consstent wth the hypothess that enrollees dscount ther MSAs. The smaller ther MSAs are, the hgher the chance of payng medcal expendtures out-of-pocket (the "true" prce): when forced to pay the "true" prce of medcal servces, they consume less. Keywords: Medcal Savngs Account, Natural Experment, Medcal Expendtures, Health JEL classfcatons: I11; I18; O12 Maoyong Fan s an assstant professor n Economcs Department at Ball State Unversty. Zhen Le s an assstant professor n Department of Energy and Mneral Engneerng at Penn State Unversty. Guoen Lu s a professor n Guanghua School of Management at Pekng Unversty.
1. Introducton The reform n urban health care system s among the most sgnfcant polcy reforms n Chna snce the 1990s. The tradtonal health care system, establshed n the 1950s and pllared by the government nsurance system (GIS) and the labor nsurance system (LIS), had ran nto fnancal dffcultes snce the late 1980s, due to escalatng healthcare cost and mnmal consumer costsharng (Wnne C. Yp and Wllam C. Hsao, 1997). The reform was ntated n 1994 n two plot ctes, Zhenjang and Jujang, was extended to 57 other ctes n 1996, and by the end of 1998, a natonwde reform campagn was carred out. Whle there were certan varatons n polcy desgn across ctes the reform n all regons followed the government gudelnes and share common key features: ensurng access to basc care; wde coverage; and an nsurance polcy featured wth ndvdual medcal savngs accounts (MSAs) and a socal poolng account (SPA) fnanced by jont contrbutons from employers and employees (Gordon G. Lu et al., 2002). The reform marked Chna as the second naton, after Sngapore, adoptng MSA as a major tool of fnancng ndvdual's health care. The urban health nsurance system n Chna s n essence a MSA combned wth hgh deductble nsurance plan. MSAs are personally owned accounts, funded by contrbutons from both employees and employers. These accounts are used to pay for certan medcal expenses specfed by s nsurance polces. SPA kcks n after a certan hgh deductble s met and covers a large share of health care costs. Consstent wth arguments by MSA advocates n other countres, 1 the purpose of MSAs, accordng to Chnese central and local governments, s three-fold: (1) to contan health expendtures by addressng enrollees moral hazard and controllng ther demand for medcal servces; (2) to acheve equty by ensurng a certan level of health servce consumpton n 1 Advocates of MSA have suggested that MSAs wll reduce moral hazard, ncrease choce, mprove the effcency of nsurance, and result n lower costs and expendtures on health care.(anna Dxon, 2002). 1
medcal servces for everyone; 2 and (3) to serve as a mandatory savng tool, partcularly for those young and healthy. Whether or not MSAs can help acheve these goals depends on how enrollees vew money n ther MSAs. Deber et al. (2004) suggested that "dfferent rules [of MSA] wll clearly lead to dfferent ncentves for varous players." Emprcal studes on MSAs coupled wth a hgh deductble plan n the USA showed that people treat payments wth MSAs as less costly compared to payments wth out-of-pocket money, due to the fact that contrbutons to MSAs are tax deductble. Consequently, MSAs mtgate the effects of a hgh-deductble plan, as the reducton n medcal spendng by a person wth a MSA would be less than the reducton wth a hgh deductble plan alone. Wth regards to MSAs n Chna, there are several nterestng features that are noteworthy. Contrbutons to a MSA by an employee are mandatory and drectly wthheld from her payroll. The money n her MSA earns lttle nterest, s restrcted to pay for her medcal expendtures only, not for medcal expendtures by her household members or for her other consumpton expendtures. In prncple, the MSA money cannot be cashed out durng the enrollee s lfetme. When the enrollee des, money n her MSA wll be transferred to her her s MSA account. Wth such mandatory and restrctve nature of MSA, 3 a natural queston s rased: Do enrollees dscount the MSA money and have ncentves to use t up prematurely, just as people sgnfcantly dscount gft cards and n-knd publc assstance such as food stamps? 4 If enrollees 2 Contrbuton to an enrollee s SPA s n general based on her last year annual salary, wth certan mnmal level. In Zhenjang, for nstance, 60% of the cty average worker salary s the mnmal contrbuton base. For an enrollee whose last year annual salary was below, her contrbuton base was adjusted to that mnmal base, 3 Employee contrbuton to MSA s tax deductble as s MSA contrbuton n U.S. But snce hstorcally Chnese pay lttle attenton to ncome tax n general, we expect the tax deductblty alone does not render enrollees n Chna to dscount MSAs as much as n the U.S. 4 See Robert A. Mofft (1989) and Dane Whtmore (2002). 2
thnk MSA money has less value than money n the pocket, the purported functon of MSAs as nstruments of controllng medcal demand and of savng would be questonable. 5 Ths paper studes whether money n MSAs s sgnfcantly dscounted by enrollees n Zhenjang, one of the frst two ctes n Chna that ploted urban health nsurance reform. Under the MSA-Deductble-SPA system n Zhenjang, an enrollee pays her medcal expendtures, both for outpatent servces and for npatent servces, usng her MSA frst. After exhaustng her MSA, she s responsble for a deductble equal to 10% of her salary from the prevous year wth her out-of-pocket money. The SPA kcks n after the deductble s pad. At frst thought, t seems that money s a MSA wll be treated no dfferently than out-of-pocket money under a three-ter (MSA-deductble-SPA) pay scheme. Yet, ths s not the case. We explore a natural experment caused by a polcy change on MSA contrbuton rates n Zhenjang. Before 2002, the contrbuton rate to MSA for an enrollee who was 44 years old or younger was 4% of her last year annual salary or 60% of the average cty worker salary, whchever hgher. In late 2001 the cty government decded to reduce MSA contrbuton rate from 4% to 3% for enrollees younger than 35 years old, but kept the 4% contrbuton rate for those between 35 and 44. The polcy change, promulgated n December 27, 2001 and n effect n 2002, caused an exogenous shock n amount contrbuted to MSAs and thus the ntal amount n MSAs at the begnnng of 2002 for enrollees younger than 35 years old, but not for those of 35 years or older. We take advantage of ths dscontnuty and compare medcal expendtures between two consecutve brth cohorts: the cohort 1969 vs. the cohort 1968, who were 33 and 34 years old, respectvely, n 2002. These two cohorts are very smlar and thus, concerns about heterogeneous shocks n health stuaton can be assumed to be mnmal. The cohort 1969 (treatment cohort) 5 One anecdotal story s that people sometmes buy over-prced tolet paper n drug store n Zhenjang. 3
experenced an exogenous reducton n ther MSA contrbuton n 2002, whereas the cohort 1968 (comparson cohort) dd not. We employ the Dfference-n-Dfference method to estmate the mpact of the exogenous shock n MSA contrbuton on medcal expendtures. Our results show that a reducton n MSAs caused the cohort 1969 to reduce ther medcal expendtures n 2002, relatve to the comparson cohort 1968. The drop n medcal expendtures by the cohort 1969 was bgger n magntude than the amount of reducton n MSA contrbuton, and thus was unlkely caused by a mere ncome effect. 6 Moreover, we fnd that the reducton n medcal expendtures n 2002 by the treatment cohort was partcularly sgnfcant for those lkely to enter the deductble stage and pay medcal servces wth ther out-of-pocket money. Our results suggest that enrollees n Zhenjang, at least for cohorts around 35, studed n the paper, dscounted money n ther MSAs sgnfcantly. When ther MSA accounts were reduced, enrollees, Ceters Parbus, were more lkely to pay ther medcal servces out-of-pocket. Therefore, when the ntal amount n MSAs was reduced, the "true" cost of ther medcal servces, the cost they really cared about, was lkely to be hgher. As a result, they tended to cut ther medcal servce consumpton n response to a reducton n ther MSAs. The fndng that enrollees n Zhenjang sgnfcantly dscounted money n ther MSAs and dd not vew them as ther own money, suggests that we need to be cautous about usng MSAs as an nstrument to address enrollees moral hazard and contan medcal expendtures and to encourage enrollees to save. Interestngly enough, despte sgnfcant dfferences n the MSA rules between Sngapore, the frst naton that adopted MSAs, and Chna, some studes on the 6 The ncome effect here refers to the possblty that the cohort 1969 s total dsposable ncome was reduced, relatve to the cohort 1968, f the total ncome ncluded money n the MSA accounts. But an ncome effect s unlkely to cause a reducton n medcal expendture that s even bgger than the reducton n the ncome. 4
Sngapore experence concluded that "there s no evdence that they [MSAs] have been effectve n restranng health costs" (Mchael D. Barr, 2001). The nsghts whch our study provdes on how enrollees n Zhenjang vewed and treated money n ther MSAs suggest an nterestng perspectve on the role of MSAs n health nsurance. The study also hghlghts the mportance of detals and rules of MSA desgn f a country s to ncorporate MSAs n the health nsurance polcy. The rest of the paper s arranged as follows. Secton 2 dscusses wde spread use of MSAs n urban health nsurance reform n Chna and MSAs n Zhenjang cty n partcular. Secton 3 presents our hypothess that MSAs are dscounted sgnfcantly by enrollees n Zhenjang, and a theory underlyng our hypothess testng that shows how a reducton n a MSA would have a dfferent mpact on an enrollee s medcal expendtures f she vews or does not vew money n her MSA as her own money. Secton 4 dscusses data, followed by the emprcal strateges n secton 5. Secton 6 presents estmaton results. Secton 7 concludes. 1. Chnese Urban Health Care Reform and MSA 2.1. MSA-Deductble-SPA System n Chna Pror to reform, the urban health care system n Chna had been fnanced prmarly through two major publc programs: the Government Insurance Program (GIP) and the Labor Insurance Program (LIP). The GIP was prmarly for government employees, veterans, educators, and college students; whereas the LIP was for workers of all state-owned and some non-state-owned enterprses. There were two major problems wth the old system. Frst, both consumers and provders of health care had ncentve to abuse the systems. 7 Second, there was no rsk poolng 7 On one hand, enrollees receved outpatent and npatent medcal servces wth mnmal cost sharng so that they had no ncentve to seek the most cost-effectve health care. On the other hand, hosptals were usually rembursed on 5
across workng unts (enterprses or government agences) because each workng unt was selfnsured. 8 Consequently, the urban health care system reached a major health crss due to rapd cost escalaton and ncreasng nequalty n health care fnancng, leadng to the urban health nsurance reform that started wth plot experments n Zhenjang and Jujang n 1994 and culmnated wth a natonwde mplementaton n 1998. The reform has brought about three mportant changes. Frst, t mandates all enterprses and government agences n the communty (usually a cty) to partcpate n a sngle-payer and cty-run health nsurance plan. Ths serves three prmary functons: mnmzng the rsk of selecton bas (both adverse selecton by patents and favorable selecton by provders); maxmzng the poolng of health rsks; and enablng equtable health care access. Second, the reform mplements a combnaton of MSAs and SPA to fnance medcal expendtures. MSAs are owned by ndvdual enrollees and the SPA s shared by all. MSAs and SPA are fnanced through jont contrbutons of employers, employees and local government subsdes. Thrd, the nsurance benefts are comprehensve, coverng major servces ncludng npatent care, outpatent care, emergency room (ER), and medcatons, coupled wth varous demand and supply sde cost sharng mechansms n an attempt to control costs. All partcpatng provders obtan rembursement for servces drectly from the Cty Socal Securty Bureau. The rembursement rates are determned accordng to fee schedules varyng wth the level and type of provder. a fee-for-servce bass accordng to a government-set fee schedule, whch gave provders ncentve to over-provde servces. 8 No rsk poolng under the old system was no problem pror to Chna s economc reform, as all state-owned enterprses were guaranteed the basc health care benefts through government subsdes. The economc reform detached most of the enterprses from the central government fnancally. As a result, those non-proftable stateowned enterprses had lttle capacty to remburse large medcal blls for ther employees and retrees, who thus were effectvely unnsured and had to pay out-of-pocket for ther health care costs. Ths causes fnancal hardshps for the elderly and those wth chronc dseases. On the other hand, those proftable enterprses could contnue to provde generous health care benefts, whch often contaned lttle patent cost-sharng responsbltes. 6
Across ctes and regons n Chna, the MSA-Deductble-SPA system has taken two major forms, whch are dstngushed by rules on whch servces are pad by whch accounts. One form s often referred to as the Ppelne Model, or the Zhenjang Model, as t was frst mplemented n Zhenjang and Jujang, the two plot ctes n the reform. In the Ppelne Model, no dfferentaton between outpatent servces and npatent servces s made. An enrollee uses her MSA to pay her medcal expendtures frst, ether outpatent or npatent, untl all MSA money s spent. Then, a deductble has to be pad out-of-pocket before the SPA kcks n. The SPA pays a majorty of her medcal expendtures beyond the deductble, wth a certan percentage of copayments pad by the enrollee. The other form s called the Mxed Model, or the Nanjng Model. 9 In the Mxed Model, the SPA covers an enrollee s outpatent expendtures for chronc dseases and the non-copayment porton of npatent servces beyond a certan hgh deductble. The enrollee can use her MSA money to pay for outpatent servces, the deductble and the copayment porton of npatent servces she receves. The combned system of MSA and SPA n Chna, n ether model, n essence conssts of a MSA plus a hgh deductble nsurance plan. MSAs are used to contan medcal expendtures through demand control, ensure a certan level of equty n terms of access to basc medcal servces, and serve as mandatory savngs, prmarly for those young and healthy. 2.2. MSAs n Zhenjang Zhenjang adopted the Ppelne Model n 1994, whch conssts of a three ter pay scheme: MSAs, deductbles and the SPA. It s mandatory for employed people to partcpate n the cty nsurance program and the cty authorty creates a personal MSA for each enrollee. MSAs are funded 9 The Mxed model has evolved from an ntal model for whch outpatent servces are pad frst by MSAs and then by out-of-pocket money, whereas npatent expendtures beyond a hgh deductble are pad by SPA wth copayments. 7
jontly by employers and employees. The contrbuton rate vares for ndvduals wth dfferent ages and dfferent workng statuses. Fgure 1 llustrates contrbuton rates by employees and employers for those younger than 44 years old n 2001 n Zhenjang. Both the enrollee and her employer make contrbutons to the enrollee s MSA account, wth the employee contrbutng 2 percent of her last year annual salary (a.k.a. contrbuton base) and the employer contrbutng 2 percent of the contrbuton base. The contrbuton by the enrollee s drectly wthheld from her payroll on a monthly bass. The employer, n addton, contrbutes 6% of the employee s base to the SPA on her behalf. A self-employed ndvdual can partcpate n the program by contrbutng 4% of the cty s last-year average annual salary to her MSA account and the SPA, respectvely. When medcal expendtures are ncurred, whether for outpatent or npatent servces, the MSA s used frst to cover these expenses. After the enrollee exhausts her MSA, she then must pay out of pocket untl her deductble s reached. The deductble s a certan percentage of her contrbuton base. After the deductble s met, the SPA kcks n to cover the remander of the cost. Even wth the SPA, the enrollee s stll responsble for a copayment for each medcal servce beyond the deductble. The copayment percentage vares accordng to whether t s an npatent or outpatent servce, who provdes the servce, and how much the expendture s. 10 There s also an upper lmt of 3,500 RMB on the total annual amount of copayments that an enrollee pays, beyond whch the SPA wll cover any remanng cost. Moreover, an upper lmt of 30,000 RMB s set for the total annual amount of SPA and copayments. Beyond ths lmt, another nsurance plan wll kck n. 11 10 For a workng enrollee, the copayment rates for an npatent servce are progressvely reduced: 20% for the bracket of RMB 1-5000, 10% for the bracket of RMB 5001-10000, and 5% for RMB 10001-30000. A retred enrollee s copayment rates are half of those for a workng enrollee. The contrbuton rates for an outpatent servce are 35% f the servce s provded by a Level 3 hosptal (cty owned hosptals, etc), 30% f a Level 2 hosptal, and 25% f a Level 1 hosptal (communty clncs, etc). 11 Ths nsurance plan s also mandatory and ams to cover npatent servces that are very costly. The annual premum s RMB 60 for workers and RMB 48 for retrees. 8
Not only are MSAs n Zhenjang mandatory, but ther use s restrctve. An enrollee s MSA can be used to pay for her medcal expendtures only; she cannot use her MSA to pay for health care costs of her famly members or for other consumpton. If her MSA account s not exhausted at the end of a year, the surplus wll be retaned n her MSA for next year and cannot be taken out. Interest s pad on the balance of her MSA at the end of the year at the current annual nterest rate of a bank checkng account. The nterest s also kept n the MSA account. Durng the enrollee s lfetme, money n MSA can not be cashed out. When the enrollee des, her MSA wll be transferred to her her s MSA account f the her has one. If the her does not partcpate n the plan, her MSA can be used to pay for the her s medcal expendtures untl exhauston. If the enrollee has no her, when she des, what s left n her MSA wll be transferred to SPA. 2. Hypothess and A Smple Theory 3.1. Hypothess Offcal documents from Bureau of Health Insurance n Zhenjang have repeatedly stated two man purposes of MSAs: to contan fast growng medcal expendtures and to encourage young and healthy people to accumulate funds for future use. Both purposes are hnged on a crucal assumpton that enrollees treat money n MSA the same as out-of-pocket money. If ths assumpton holds, MSAs can functon as an addtonal layer of deductble to reduce enrollees moral hazard. However, ths paper hypotheszes that enrollees n Zhenjang sgnfcantly dscount money n MSAs and do not treat them as valuable as out-of-pocket money as MSAs are compulsory and restrctve. Even though contrbutons to MSAs are shared by employers and 9
employees, employees have no drect control over ther contrbutons; the mandatory MSA contrbutons functon lke a payroll tax. Money n ther MSAs s restrcted to pay for ther own medcal expendtures and cannot be cashed out. As a consequence, money n MSA could seem less valuable for enrollees than cash-on-hand. Therefore, enrollees mght feel that t s not necessary to constran medal expendtures untl they exhaust ther MSAs. If ths s the case, the MSA wll not functon as the polcy desgners n Zhenjang had thought and hoped. How do we emprcally test our hypothess that enrollees n Zhenjang dscount money n ther MSAs sgnfcantly? We need an exogenous shock n ndvdual MSAs n order to examne how enrollees responded to the shock. If the MSA money s as valuable as out-of-pocket money, we should not expect any change n behavor. However, f the MSA money s dscounted, a reducton n ndvdual MSAs wll rase the "true" prce of medcal servces and cause ndvduals to reduce ther demand. The followng paragraph presents the ntuton underlyng our hypothess. Consder an enrollee for whom there exsts some probablty that she wll exhaust her MSA account and enter the deductble stage durng the comng year. When she experences a reducton n her MSA at the begnnng of the year, Ceters Parbus, the chance of her to enterng the deductble stage where she pays her medcal expendtures wth her out-of-pocket money ncreases. If she dscounts money n her MSA sgnfcantly, not vewng t as beng as valuable as out-of-pocket money, then what she really cares about s the out-of-pocket payment durng the deductble stage, whch s the "true" cost to her. Therefore, wth a reducton n the ntal money n her MSA, the "true" cost of her medcal servces, other thngs beng equal, s expected to become hgher. Gven that medcal servces are normal goods, we would expect her to reduce her consumpton. 10
Consequently, her medcal expendtures durng that year wll drop n response to a reducton n the ntal amount of her MSA. However, f the enrollee vews money n her MSA the same as out-of-pocket money, a reducton n her MSA at the begnnng of the year wll not change the "true" cost of her consumpton of medcal servces, unless she enters the thrd stage where SPA kcks n.. In ths case, a reducton n her ntal MSA amount would have lttle mpact on her medcal expendtures. Furthermore, f t s lkely for the enrollee to enter the SPA stage, then a reducton n her MSA account could render the probablty of enterng the SPA stage even hgher and the expected "true" cost of her medcal consumpton lower, as the "true" cost of her medcal servces s small when SPA covers most of the expendtures. As a result, the enrollee s demand for medcal servces could ncrease n response to a reducton n the ntal MSA amount. Therefore, how an enrollee s medcal expendture responds to the change n the amount of her MSA could reflect how she vews money n her MSA. Ths s the dea underlyng our emprcal test of whether an enrollee consders her MSA money the same as cash-on-hand or f she dscounts t sgnfcantly. 3.2. Graphcal llustraton Fgure 1 shows, wth a numerc example, the relatonshp between "true" cost and nomnal medcal expendture for an enrollee who does not value money n her MSA at all, an extreme example of dscountng. Suppose that the enrollee has a contrbuton base (her last year annual salary) of 10000 RMB, the ntal amount of money n her MSA s 400 RMB, and a deductble of 1000 RMB. The relatonshp between the enrollee s "true" cost vs. her nomnal medcal expendture s shown as lne OABC. For the frst 400 RMB of her medcal expendture, the 11
"true" cost s zero. When MSA s exhausted, the enrollee enters the deductble stage and has to pay out-of-pocket up to the 1000 RMB deductble lmt. The slope of AB s 45 degrees. When the deductble lmt s reached, the SPA kcks n and pays a majorty percentage of her further medcal expendtures. The slope of BC depends on the copayment rate for the enrollee durng the SPA stage. When the ntal MSA amount s reduced from 400 RMB to 300 RMB, the lne of "true" cost to the enrollee vs. nomnal medcal expendture shfts from OABC to OA B C. It s clear to see that a reducton n MSAs has dfferent mpacts on dfferent types of enrollees. For an enrollee who s unlkely to enter the deductble stage, mostly due to a good health status and thus a low medcal expendture (left part), a reducton n the MSA amount at the begnnng of a year s unlkely to change the "true" cost of her medcal servces and thus to change her level of medcal servce consumpton. For those who are lkely to enter the deductble stage, because of ther ntermedate or hgh medcal expendtures, a reducton n the ntal MSA amount causes them to face a hgher average prce for medcal servces. If health care servces are normal goods, the demand for medcal servces wll decrease and ther medcal expendtures durng that year wll drop. Moreover, t s shown n the fgure that a reducton n a MSA has much bgger mpact on enrollees who have ntermedate medcal expendture and are lkely to enter the deductble stage but unlkely to enter the SPA stage (mddle part) than enrollees wth hgh medcal expendture (far rght part). For enrollees wth hgh medcal expendtures (who use the SPA), health care servces only become slghtly more costly and we do not expect ther medcal expendtures to respond to a reducton n MSAs as much as those of enrollees wth ntermedate level of medcal expendtures. Therefore, f enrollees dscount ther MSA money sgnfcantly, a reducton n the ntal MSA amount would have the followng effects: (1) people wth low 12
medcal expendtures who are unlkely to enter the deducble stage are less lkely be affected; (2) people wth ntermedate medcal expendture who are lkely to enter the deductble stage but not the SPA stage wll be affected most and wll decrease ther medcal expendtures; (3) people wth hgh medcal expendtures who are lkely to enter the SPA stage wll also decrease ther medcal expendtures, but the drop s lkely to be small. Fgure 2, by contrast, shows the other extreme case where an enrollee consders money n a MSA to have the same value as cash-on-hand. Spendng wth MSA s thus no dfferent from spendng usng cash. Lne OAB overlaps the 45 degree lne, and there s no knk at pont A of enterng the deductble stage. One RMB medcal expendture amounts to one RMB "true" cost to the enrollee. The SPA kcks n when the deductble lmt of 1,000 RMB s reached. When the ntal MSA amount s reduced from 400 RMB to 300 RMB, the lne OBC shfts to OB C. From the fgure, the "true" prce of health servces does not change for enrollees wth low or ntermedate medcal expenses who are unlkely to enter the SPA stage, but enrollees wth hgh medcal expendture now faces an average lower "true" prces. Therefore, f enrollees do not dscount money n ther MSAs, ther medcal expendtures, n response to a reducton n MSAs, s lkely to reman unchanged for those wth low or ntermedate medcal expendtures, but to ncrease for those wth hgh medcal expendtures. 3.3. A Smple Model We use a smple model to llustrate that f an enrollee does not value the money n her MSA at all, her decson on how much medcal servces to consume wll be postvely dependent on the ntal MSA amount. The enrollee cannot predct accurately her medcal expendtures durng a 13
year, whch s stochastc and depends on her health status and the nature. 12 But she has knowledge about the dstrbuton of her medcal expendtures, X, whch s the total annual expendtures f all her medcal needs are satsfed. Let the dstrbuton of X follow an exponental dstrbuton: (1) λx λe, x 0 f( x; λ) = 0, x < 0 The parameter λ represents the health status of the enrollee. A larger λ ndcates better health. Let α be the probablty of seeng a doctor when a enrollee s sck, and the range of α s [0,1]. 13 Gvenα, the enrollee s total medcal expendtures would be Xα. For smplcty, let us assume that the SPA stage does not exst and the enrollee pays her medcal expendtures frst wth her MSA and then out-of-pocket. Let M represent the ntal amount n her MSA n the begnnng of a year. Gven that the enrollee does not value her MSA at all and that the probablty of her seekng medcal servces sα, the expected "true cost" of her medcal consumpton durng the year s: (2) ( α ) M λx α M x M e dx e λ α λ = α λ However, there s a dsutlty assocated wth not seeng a doctor when sck. Let the total dsutlty of not seekng for medcal servces when needed durng the year s: (3) 1 Dsutlty = C( 1) X α 12 For example, the outburst of flu or a car accdent. 13 For nstance, f the number of tmes the enrollee beng sck s N 1 n a year and the number of tmes that the enrollee goes to see a doctor s N 2, then α can be consdered to be N 2 /N 1. 14
where C s a postve and ndvdual specfc coeffcent. The cost s hgher when the total medcal expendture X s hgher or the probablty of gong to see a doctor when needed, α, s lower. Expected dsutlty for the enrollee s: (4) 0 1 λ C 1 ( 1) x C xλe dx = ( 1) α λ α At the begnnng of the year, the enrollee decdes α to mnmze the sum of her expected "true" medcal cost and the expected dsutlty: (5) Mn α M α λ C 1 e α + ( 1) λ λ α Take frst order condton w.r.t. α, we get: (6) M e λ α M 2 C 1+ λ = 0 α λα The relatonshp between α, the probablty of gong to see a doctor when sck and the ntal MSA balance M s: (7) M λ 2 λ αme α M λ 2 2 λ M e α + 2C α = > 0 M It says that when the ntal MSA balance s reduced, the probablty of seeng a doctor becomes lower for the enrollee. Snce total medcal expendture equals Xα, the total medcal expendtures α α wll also be lower. Moreover, when λ, 0. Ths ndcates that very healthy M M people do not respond to the ntal MSA amount. 15
We do a smple comparatve statc analyss to show whch enrollees respond most to a shock n ther MSA n terms of ther decsons to see a doctor when sck, and as a consequence, ther medcal expendtures change. We take frst order dervatve of equaton (7) w.r.t. λ : (8) M λ C M M e α = M 2 λ λ 2 2 λ M e α + 2C 2 2 ( α M ) 2 ( 2λα λ ) ( ) α M Solve for λ = 0, we get λ = α * 2 M *. And we plug λ back n equaton (7), we get: α M > 0 * 2α λ = M. Thus, enrollees wth λ = α * 2 M respond most to a change n the ntal MSA balance, M. When M s reduced, ther α and annual medcal expendtures are reduced most * dramatcally. When λ > λ, ( α M ) λ < 0. It says that enrollee s less responsve to change * n MSA when he becomes healther than λ. When λ < λ, * ( α M ) λ > 0. It says that enrollee s less responsve to change n MSA when he becomes less healthy than λ *. Therefore, ths smple model llustrates that f enrollees dscount ther MSA money sgnfcantly, a reducton n the ntal MSA balance wll reduce ther medcal expendtures. Such effects are the bggest for enrollees wth an ntermedate health status. 3. Data and Emprcal Strategy 16
To emprcally test how enrollees medcal expendtures respond to a reducton n MSAs, we explore a natural experment that caused an exogenous shock to MSAs for certan brth cohorts but not others n Zhenjang. The cty authorty adjusted the MSA contrbuton rates for dfferent age cohorts n December 2001. The new polces took effect on January 1, 2002. Before the polcy change, the requred contrbuton to MSAs s 4% of ndvdual annual salary for all enrollees younger than 45 years old. The new polcy cut the contrbuton for enrollees younger than 35 years old from 4% to 3% whle mantanng the rates at 4% for people who were between 35 and 45 years old. The polcy change creates an exogenous shock n MSA contrbutons for those born n 1969 or later. Ther MSA contrbutons suffer from a sudden loss of 1% of ther annual salary. For those born n 1968 and before, ther MSA contrbutons do not change. Snce brth cohorts 1968 and 1969 should be very smlar n terms of health and health care demand, we can compare changes n ther medcal expendtures before and after the polcy change and attrbute the dfferences to MSA levels. We summarze the polcy change n Table 1. There are four brth cohorts n the table: 1968, 1969, 1970 and 1971. The contrbuton rate for all cohorts was the same, 4%, n 2000 and 2001. But n 2002, due to the polcy change, the MSA contrbuton rate dropped to 3% for two younger cohorts, 1970 and 1971, but remaned at 4% for two older cohorts, 1968 and 1969. 4.1. Data Data for ths study s from Zhenjang's socal health nsurance database managed by the cty government. The database was establshed to montor medcal expendtures of enrollees covered by Urban Employee Health Insurance Program (UEHIP). The data set covers every employed or 17
retred person n the cty. We focus on a panel of four cohorts, 1968, 1969, 1970 and 1971, whch were contnuously enrolled n the UEHIP from 2000 to 2003. We exclude employees workng for government agences because they were subsdzed by other sources of health care funds for whch we have no nformaton. Snce the UEHIP covers all enterprses (both stateowned and prvate) n Zhenjang and each worker has to partcpate, we do not worry about the entry-ext type of attrton problem whch s common n most health nsurance datasets. 14 The data set records every enrollee's annual salary whch s used to determne the amount of ndvdual MSA contrbutons each year. Indvduals total annual medcal expendtures are also recorded. We also know the balance of MSAs at the begnnng of each year. Wth ths nformaton, we can easly tell whether or not enrollees' medcal expendtures reach the deductble or SPA levels. Other demographc nformaton s lmted except age, gender, occupaton, and retrement status. 4.2. Dfference-n-Dfference (DID) Model Snce we have panel data, we employ the panel Dfference-n-Dfference (DID) method to estmate the average effects of a reducton n MSA on medcal expendtures. We compare the change n medcal expendtures from 2001 to 2002 for cohorts that experenced a cut n ther MSAs and cohorts that dd not. Specfcally, we run the followng model: Y = α + γtg TY + λty + X β + φ + ϕ +ε (9) t t t t t t Where Y s enrollee s total medcal expendtures n year t. TG s a dummy varable wth 1 t ndcatng the treatment group. TY t s a dummy varable wth 1 ndcatng the year of 2002. Snce 14 Based on offcal documents, more than 90% of all employed people were enrolled n UEHIP n Zhenjang from 2000 to 2002. 18
we estmate equaton (9) usng the fxed effect model, TG dummy drops out. X s a vector ncludng enrollee s annual salary and left-over money n her MSA from last year n year t, both of whch are tme-varyng control varables. φ s the ndvdual fxed effects whch captures the dfferences of demand for medcal care due to ndvdual specfc tme-nvarant factors lke genetcs. ϕ t s the year fxed effects whch captures. ε t s the error term. Moreover, our theory predcts that enrollees wth ntermedate health status who are lkely to enter the deductble stage are most lkely to be affected by a reducton n MSAs. Therefore, we dvde the cohorts nto three subgroups accordng to ther level of medcal expendtures before the treatment. Then we run the followng equaton to allow the treatment effects to vary across the three subgroups: t (10) Y = α + γ TG TY SG + γ TG TY SG + γ TG TY SG 1 2 t 1 t 2 t 3 t + λ TY SG + λty SG + λ TY 3 SG + X β + φ +ϕ t +ε 1 2 3 1 t 2 t t t 3 t j 1 where SG ndcates that enrollee belongs to a certan subgroup j. SG refers to the subgroup 2 of enrollees who were relatvely healthy and unlkely to enter the deductble stage SG refers to the subgroup wth ntermedate health status who are lkely to enter the deductble stage but not the SPA stage. 3 SG ndcates the subgroup whose total medcal expendtures before the treatment year were above the SPA payment lne. The key coeffcents are parameters to three thrd-order nteractons n equaton (10). They ndcate how enrollees n each of the three subgroups react to a shock n MSA contrbuton. 4.3. Matchng Gven such a large number of enrollees n our data, we also employ covarate matchng (CVM), a sem-parametrc econometrc method wth the advantage that no specfc parametrc 19
relatonshp between the outcome and explanatory varables needs to be assumed. CVM compares an ndvdual n the treatment group wth ndvdual(s) n the comparson group drectly based on ther ndvdual characterstcs. In CVM, every treated unt s matched to a number of unts n the comparson group based on the dstance measured by the vector norm. Let ( ) 1/2 x V = xvx ' be the vector norm wth postve defnte matrx V 15, the CVM defnes z x as the dstance between the vector x and z, where x and z represent the covarates for a V treated unt and a potental match. Let d M () be the dstance from unt to the M th nearest match wth the opposte treatment. Consder the set of observed covarates for unt to be unt wll match wth s: { M } () l 1,, N T 1 T, X X d () Ψ = = = M l l V X, the set and d M () s defned as lt : l= 1 T 1 { X l X dm () } =M V where 1{} s the ndcator functon, whch s equal to 1 when the value n brackets s true and zero otherwse. The smple matchng estmator wll be based n fnte samples when the matchng s not exact. Abade and Imbens (2002) develop a bas-corrected matchng estmator adjustng the dfference wthn the matches for the dfferences n ther covarate values. Although theoretcally 15 We use the dagonal matrx, of whch the dagonal elements are the nverses of the varances of X (the element of the set of covarates), as our weghtng matrx V. The weghtng matrx V accounts for the dfference n the scale of the covarates. 20
matchng on multdmensonal covarates can lead to substantal bas, the matchng approach combned wth bas adjustment often leads to estmates wth lttle remanng bas. 5. Results 5.1. Effects on Total Medcal Expendtures Frst, we want to show that the treatment cohort(s) and the comparson cohort(s) are comparable. Presumably, ther medcal expendtures would be smlar n the absence of the treatment. Therefore, any dfferences between those two groups attrbute to the treatment. Table 2 shows means and standard devatons of man characterstcs of two groups, ncludng ther annual salary, MSA contrbutons and ther varous medcal expendtures n 2000 and 2001. Smple t- tests show that those two cohorts are statstcally ndfferent. The average contrbuton to MSA, the total medcal expendtures, and the total amount of money n MSA over tme are plotted n Fgure 4. Sold (dashed) lnes are for the cohort 1969 (1968). It can be seen clearly from the graph that cohorts 1969 and 1968 are very smlar to each other n terms of contrbutons to MSAs, ntal MSA balances, and medcal expendtures n 2000 and 2001. However, all three measures dverged n 2002. For the cohort 1969, total medcal expendtures n 2002 decreased whle both MSA contrbutons and ntal MSA amounts dropped due to the polcy change. Table 3 presents the estmated results of DID. Column 1 shows DID estmates for two brth cohorts 1969 and 1968. Panel A presents estmates of the average effect of the polcy for the whole group, estmated from equaton (9). The results show that a reducton n MSAs of the treatment group reduced ther total medcal expendtures sgnfcantly n 2002. Wth 1% 21
reducton n MSA contrbuton rate, 16 amountng to an average reducton of 87 RMB n ther ntal MSA accounts at the begnnng of 2002, the cohort 1969 reduced ther annual medcal expendtures by 106 RMB. Our theory predcts that enrollees wth medcal expendtures reachng the deductble level wll react to a reducton n MSA most sgnfcantly. Based on our theory, we dvde the full sample nto three subgroups. The frst group ncludes enrollees that never exhausted ther MSA n both 2000 and 2001. The thrd group ncludes enrollees that entered SPA at least once n 2000 and 2001. The rest belongs to the second group whch s mostly lkely to be affected by the polcy change. There are 5309, 2234, and 726 enrollees n the frst, second, and thrd group. Panel B of Table 1 present the effects for these three subgroups, based on estmates of equaton (10). We can see that the coeffcent s sgnfcant and negatve for the second group, but not for other two groups. Consstent wth our theory, those enrollees whose medcal expendtures are more lkely to enter the deductble stage are most senstve to the polcy change. In Panel C of Table 1, we dvde the full sample nto three subgroups accordng to ther level of total medcal expendtures before 2002, the treatment year. We use total medcal expendtures as a nosy proxy for health status, and thus the estmates could provde us useful nformaton concernng the effects of MSAs on enrollees wth dfferent health statuses. Lower Quntle ndcates that enrollee's average annual medcal expendture n 2002 and 2001 falls nto 0-33.3 percentles, Mddle Quntle 33.3-66.7 percentles, and Upper Quntle ndcates the rest of enrollees. Our results show that, for those wth low medcal expendtures, a 25% reducton n MSA contrbuton has no sgnfcant mpact on total medcal expendtures. For those wth medcal expendtures n the mddle, a 25% reducton n MSA contrbuton led to statstcally 16 Gven that prevous contrbuton rate s 4%, a change from 4% to 3% equals 25% reducton n the amount of contrbuton. 22
sgnfcant cut n total medcal expendture. Fnally, for the group wth hgh medcal expendtures, the coeffcent s negatve and sgnfcant at the 5% level. Column 2 shows DID estmates for a larger sample. The treatment ncludes cohorts 1969 and 1970 and the comparson ncludes cohorts 1967 and 1968. The conclusons by and large hold. Even though the estmates become smaller, the story stll holds. Results from CMV matchng are shown n Table 4. Each entry represents an estmate of average treatment effects of a 25% reducton n MSA contrbutons on medcal expendtures for the treated subsample. To mnmze the mpact of the sze of MSA, we match the sze of MSA exactly. Other matchng varables nclude gender and annual salary of pre-treatment perod. We also experment wth dfferent number of matchng neghbors to ensure the robustness of our matchng results. We report matchng results usng 1, 5 and 10 neghbors. Panel a of Table 4 presents the number of enrollees n the treatment/comparson group and matchng results estmates for the cohort 1969 vs. the cohort 1968. Results usng dfferent number of matchng neghbors are smlar. Our results show that, for those whose medcal expendtures dd not reach the deductble stage n 2000 and 2001, a 25% reducton n MSA contrbuton has no sgnfcant mpact on medcal expendtures. For those whose medcal expendtures reached the deductble stage n 2000 or 2001, a 25% reducton n MSA led to a statstcally sgnfcant cut n total medcal expendtures. Fnally, for the group wth hgh medcal expendtures and whose medcal expendtures reached SPA n 2000 or 2001, there were no sgnfcant effects even though the coeffcent s negatve. The concluson reached wth the matchng method s consstent wth that of the DID model. Even though we have a relatvely large sample sze, we are stll concerned about outlers whch could drve the results n theory. Therefore, we dropped those enrollees who experenced 23
serous sckness and ncurred a large amount of medcal expendtures durng our sample perod. Our crteron s that anyone whose change n year-to-year medcal expendture s more than twce ther average ndvdual salary s dropped from the samples. We re-run our equatons and matchng after elmnatng the outlers. The new results are very consstent wth our man results. 17 5.2. Comparng Cohorts Usng Placebo Laws We dd two placebo law tests to ensure that the estmaton on reducton n treatment cohorts total medcal expendture was due to the exogenous shock n ther MSA contrbuton, caused by the polcy change n 2002. (Bertrand et al. 2004) Frst, we tested the dfference between cohorts that are ether both affected by the polcy or both unaffected by the polcy n 2002. Second, we use 2001 as a placebo treatment year. We then estmate equaton (9) and (10) usng the fxed effect model on these placebo laws. The estmate generates an estmate of the laws' "effect". Table 5 presents the results of the frst placebo law. We compare cohort 1969 wth cohort 1970. Both cohorts experenced a 25% reducton n MSA contrbutons n 2002. We can see that there s no sgnfcant dfference n the two cohorts change n total medcal expendtures from 2001 to 2002. We also compare cohort 1967 wth cohort 1968, for both of whch MSA contrbuton rates dd not change n 2002. Agan, there s no sgnfcant dfference between these two cohorts n terms of total medcal expendture from 2001 to 2002. Table 6 presents the results of the second placebo law, usng 2001 as a placebo treatment year. 18 Panel A shows estmates usng cohorts 1968 and 1969. Panel B shows estmates usng 17 Due to the length of the paper, those results are not presented. They are avalable upon request for nterested readers. 18 Placebo law usng 2003 as the treatment year generate smlar results. They are avalable upon request. 24
cohorts 1967 and 1968. Panel C shows estmates usng cohorts 1969 and 1970. None of these estmates are statstcally sgnfcant. 5.3. Decomposng the Effects on Total Medcal Expe1ndtures Usng Smulaton So far we have shown that wth a 1% reducton n MSA contrbuton rate n 2002, amountng to an average reducton of 87 RMB n ther ntal MSA accounts at the begnnng of 2002, the cohort 1969, on average, reduced ther total annual medcal expendtures by 105.8 RMB. In Zhenjang, an enrollee s total annual medcal expendture s the sum of three components: MSA payment, deductble payment, and the SPA payment,. How s a reducton n total annual medcal expendtures dstrbuted among the three components? Our theory predcts that wth a reducton n a MSA amount, an enrollee has a tendency to cut her demand for health servces; gven the stochastc nature of her demand for medcal servces, each of the three component payments can be mpacted. A further look nto how a reducton n a treated enrollee s annual medcal expendtures s decomposed provdes not only a robust check but also further nsghts on her behavoral change n response to a MSA reducton. However, we cannot apply the DID model drectly to the recorded MSA payments, deductble payments and SPA payments n the data, to see f there s a reducton n these three component payments that s caused by the behavoral change, n response to a reducton n MSA, by the treatment cohort. A reducton n the ntal MSA amount for the treatment cohort would lead to a dfference n the recorded component payments between the treatment and comparson cohorts, even f the treatment cohort had not cut back ther medcal consumpton. To llustrate, let us go back to the numercal example n Secton 3. Suppose that enrollee A n the treatment cohort and enrollee B n the comparson cohort, who are smlar to each other, 25
both had an ntal balance n a MSA account of 400 RMB and total medcal expendtures of 700 RMB n 2001. In 2002, enrollee B stll has an ntal MSA amount of 400 RMB and total medcal expendtures of 700 RMB. The total expendture was pad usng 500 RMB from the MSA and a deductble payment of 300 RMB and recorded n the data. For enrollee A, her ntal MSA account was reduced to 300 RMB n 2002. Now suppose ths enrollee dd not reduce her medcal servce consumpton n response to the reducton n her MSA and stll had a total medcal expendture of 700 RMB. She wll pay 300 RMB usng her MSA and 400 RMB deductble. If we drectly compared the recorded MSA payments between the two enrollees, a reducton n the MSA payment for the treatment enrollee wll be found, but ths reducton s not due to a behavoral change by the treated enrollee but to an accountng matter. Therefore, to dentfy any change n the three component payments, whch was caused by a behavoral change by the treatment cohort n response to a reducton n MSA, we need to control for an accountng change. To control for such accountng dfferences, we pretend that the comparson group (the cohort 1968) also had a reducton n ther MSA contrbuton rate n 2002, from 4% to 3%, and revse ther MSA, deductble and SPA payments based on ther pretended ntal MSA amount, the deductble and ther total annual medcal expendture n 2002 that were recorded. Dong so, we create an adjusted comparson group whose behavor can be consdered unchanged but whose three component payments were adjusted for a reducton n MSAs. Therefore, any dfference between the treatment cohort 1969 and the adjusted comparson group wll pck up the behavoral change by the treatment cohort n response to the 1% reducton n MSA contrbuton rate. The results are presented n Table 7. 26
5.3.1. MSA Payment Compared to cohort 1968, cohort 1969 s MSA payments, overall, dropped by a sgnfcant 11.0 RMB n 2002, n response to a reducton n ther MSA contrbuton. If we look at the three subgroups, MSA payments by enrollees n the MSA subgroup s reduced by 13.2 RMB, at 10% level, suggestng that some enrollees n the MSA subgroup mght cut back ther medcal servce consumpton when experencng a reducton n ther MSAs, as there was some chance for them to enter the deductble stage. For the deductble subgroup and the SPA subgroup, there was no dfference between the treatment and the comparson cohort n ther MSA payments n 2002, as the chance was hgh that they would enter the deductble stage even though they reduced ther medcal servce consumpton. Also shown n the table are the results comparng consecutve cohorts 1969 vs. 1970, whch both had a 1% reducton n MSA contrbuton rates n 2002, and comparng cohorts 1968 vs. 1967, whose contrbuton rates remaned at 4% n 2002. There s no statstcally sgnfcant dfference for these comparsons. 5.3.2. Deductble Payment The deductble payment of the treatment cohort 1969 dropped by a sgnfcant 19.4 RMB n 2002, compared to that of cohort 1968. If we look at the three subsamples, the deductble payment by the deductble subgroup n the treatment cohort was reduced sgnfcantly by 54.23 RMB, compared to ther counterparts n comparson cohort 1968. There was a reducton of 8.4 RMB, sgnfcant at 10% level, for the MSA subgroup and no sgnfcant dfference for the SPA subgroup between the treatment cohort and the control cohort. Agan, results that compare consecutve cohorts 1969 vs. 1970 and cohorts 1968 vs. 1967 show clearly that there s no statstcally sgnfcant dfference n terms of deductble payments. 27
5.3.3. SPA Expendture For SPA payment, cohort 1969 had an overall drop of 76.8 RMB n 2002, sgnfcant at 10% level. SPA payment by the SPA subgroup n the treatment cohort was reduced sgnfcantly by 122.32 RMB, compared to ther counterpart n the comparson cohort 1968. There were no sgnfcant dfference n SPA payments for the MSA subgroup and the deductble subgroup. Agan, no sgnfcant dfferences are observed, ether overall or for each of the three subgroups, between consecutve cohorts 1969 vs. 1970 and between cohorts 1968 vs. 1967. 5.4. Substtuton Effects Our results have shown that an enrollee, when faced wth an exogenous reducton n her MSA n 2002, cut back her medcal servce consumpton and thus annual medcal expendtures n that year. An nterestng queston naturally follows: dd the treatment cohort 1969, antcpatng that ther MSA contrbuton rates wll agan rse to 4% n 2003, substtute future medcal expendtures for current medcal expendtures, n the sense that some of ther medcal needs n 2002 were postponed to the next year? If so, we wll see an ncrease n ther medcal expendtures n 2003, relatve to the comparson cohort. Another possble reason for an ncrease n the treatment cohort 1979 s medcal expendtures, compared to the comparson cohort, s that enrollees who cut medcal expendtures may suffer more from the sckness n 2003 than 2002 and have to spend more on llness treatment. We test whether the cohort 1969 s medcal expendture went up from ther 2001 level, compared to the cohort 1968. Both cohorts had a 4% of MSA contrbuton rate n both 2001 and 2003. In equatons (9) and (10), the amount of MSA surplus from last year s also controlled. 28
Thus, the results wll pck up the effect of the only dfference between the two cohorts, whch s that the cohort 1969 experenced a MSA reducton and cut back ther medcal servce consumpton n 2002.Estmates are presented n Table 8. None of the coeffcents are postve and sgnfcant. Therefore, there are no substtuton effects. 6. Dscusson Medcal savngs accounts have become an mportant component of the urban health care fnancng system n Chna snce the 1990s. The prmary purpose of MSAs s to control cost by forcng patents to be responsble for ther own medcal expendtures. However, a crucal assumpton for MSAs to be functonal s that enrollees treat money n MSA the same as cash. We emprcally test the assumpton by takng advantage of a polcy change n 2002 n Zhenjang. We employ DID strategy to estmate the effects of MSA contrbuton on total medcal expendtures. We fnd that MSA contrbutons do affect patents total medcal expendtures. However, enrollees wth dfferent levels of medcal expendtures are affected dfferently. Consstent wth our theoretcal model, we fnd that enrollees wth medcal expendtures reachng the deductble stage are sgnfcantly affected, but those n MSA or SPA stages are not affected. Our fndngs have mportant polcy mplcatons. Frst, MSAs are wdely adopted n Chna s urban areas. Polcy makers expect them to contan health care costs and serve as a savngs tool. Those goals would be frutless f enrollees do not value the money n MSAs. Second, we show that compulsory MSAs based on ndvdual annual salary mght adversely affect equty n health servce usage. Snce people feel the pnch only when they need to pay deductbles, those wth small MSAs may restrct ther use of health servces even when they are necessary. 29
30
References: Abade, Abade and Imbens, Gudo W. "Smple and Bas-Corrected Matchng Estmators for Average Treatment Effects." 2002, NBER Workng Paper No. 283. Barr, Mchael D. "Medcal Savngs Accounts n Sngapore: A Crtcal Inqury." Journal of Health Poltcs, Polcy and Law, 2001, 26(4), pp. 709. Bertrand, Maranne; Duflo, Esther and Mullanathan, Sendhl. "How Much Should We Trust Dfferences-n-Dfferences Estmates?" Quarterly Journal of Economcs, 2004, 119(1), pp. 249-75. Deber, Rasa B.; Forget, Evelyn L. and Roos, Lesle L. "Medcal Savngs Accounts n a Unversal System: Wshful Thnkng Meets Evdence." Health Polcy, 2004, 70(1), pp. 49-66. Dxon, Anna. "Are Medcal Savngs Accounts a Vable Opton for Fundng Health Care?" PUBLIC HEALTH, 2002, 43(4), pp. 408-16. Lu, Gordon G.; Zhao, Z; Ca, R and Yamada, T. "Equty n Health Care Access To: Assessng the Urban Health Insurance Reform n Chna." Socal Scence & Medcne, 2002, 55(10), pp. 1779-94. Mofft, Robert. A. "Estmatng the Value of an n-knd Transfer: The Case of Food Stamps." Econometrca, 1989, 57(2), pp. 385-409. Whtmore, Dane. "What Are Food Stamps Worth?" 2002, Industral Relatons Secton, Prnceton Unversty. Yp, Wnne C. and Hsao, Wllam C. "Medcal Savngs Accounts: Lessons from Chna." Health Affars, 1997, 16(6), pp. 244. 31
Table 1. Contrbuton Rates to MSAs for Dfferent Cohorts From 2000 to 2003 Year 2000 2001 2002 2003 Cohort 1970 Contrbuton Rate 4% 4% 3% 3% Age 30 31 32 33 Cohort 1969 Contrbuton Rate 4% 4% 3% 4% Age 31 32 33 34 Cohort 1968 Contrbuton Rate 4% 4% 4% 4% Age 32 33 34 35 Cohort 1967 Contrbuton Rate 4% 4% 4% 4% Age 33 34 35 36 Notes: For each cohort, the frst row s the percentage of ndvdual annual salary and the second row s age n each year. 31
Table 2. Comparson between Cohorts 1968 and 1969 Sgnfcantly Sgnfcantly Cohort 1969 Cohort 1968 Dfferent? Cohorts 1969&1970 Cohorts 1967&1968 Dfferent? Contrbuton Base (Annual Salary) 8701.21 8758.75 N 8588.31 8681.18 N [3506.66] [3761.44] [3443.76] [3753.40] MSA Balance at the Begnnng of the Year 253.73 271.44 N 250.18 268.08 N [376.77] [399.87] [375.99] [397.67] MSA Balance at the End of the Year 366.78 370.65 N 362.56 366.58 N [156.05] [167.36] [153.22] [167.05] Medcal Expendtures n MSA Part 334.18 336.73 N 328.84 334 N [180.53] [189.60] [175.24] [188.12] Medcal Expendtures n Deductble Part 51.56 55.06 N 53.02 54.88 N [135.77] [140.42] [134.37] [139.42] Medcal Expendtures n SPA Part 156.54 162.47 N 170.21 165.88 N [994.86] [934.60] [1317.01] [1236.47] Total Medcal Expendtures 542.29 554.26 N 552.06 554.76 N [1100.83] [1046.27] [1394.44] [1328.94] # of Enrollees 3896 4373 7786 7773 Notes: Standard devatons are n brackets. Test for mean dfferences are based on t-statstcs: X Treated X Comparson t =, where n ( XTreated XControl ) 1 and n 2 are the numbers of observatons for the treatment and comparson groups. 2 2 σ σ Treated Comparson + n n 1 2 32
Table 3: The Effects of a Shock n MSA on Total Medcal Expendtures (Dfference-n-Dfference) (1) (2) Cohort 1969 vs. Cohort 1968 Cohorts 1969&1970 vs. Cohorts 1967&1968 Panel A Treatment Group Y2002-105.87* -55.74 [41.09] [30.92] Panel B MSA Treatment Group Y2002-4.35-15.79 [21.97] [17.81] Deductble Treatment Group Y2002-185.44** -120.57** [63.95] [45.38] SPA Treatment Group Y2002-662.48-221.82 [374.57] [279.49] Panel C Lower Quntle Treatment Group Y2002-2.3-10.75 [37.10] [25.88] Mddle Quntle Treatment Group Y2002-93.66** -52.31* [32.83] [25.90] Upper Quntle Treatment Group Y2002-234.05* -111.58 [110.99] [83.71] # of Enrollees 8269 15559 Notes: Robust standard errors are n brackets. * and ** ndcate sgnfcance at 5% and 1% level two tal t-test. We only report the coeffcent of nterest n the table. For the frst sample, the treatment group s the cohort 1969 and the comparson group s the cohort 1968. For the second sample, the treatment group s cohorts 1969&1970 and the comparson group s cohorts 1967&1968. The pre-treatment perod s 2001 and post-treatment perod s 2002. 33
Table 4: The Effects of a Shock n MSA on Total Medcal Expendtures (Covarate Matchng) # of Enrollees n # of Enrollees n # of Neghbors Matched the Treatment Group the Comparson Group 1 5 10 Panel A: Cohorts 1969 vs. Cohort 1968 Full Sample 3896 4373-102.81* -117.71** -117.43** [52.18] [44.76] [45.09] MSA Subsample 2525 2784-19.08-13.08-6.91 [28.77] [25.04] [23.37] Deductble Subsample 1037 1197-169.97* -219.78** -217.24** [79.85] [64.88] [60.35] SPA Subsample 334 392-741.35-889.5* -799.2* [454.43] [373.53] [359.09] Panel B: Cohorts 1969&1970 vs. Cohort 1967&1968 Full Sample 7786 7773-63.07-66.51* -74.62* [37.36] [33.36] [34.02] MSA Subsample 4897 4893-17.43-17.07-18.58 [21.85] [19.71] [18.41] Deductble Subsample 2224 2170-149.44* -172.52** -174.74** [69.95] [55.63] [53.97] SPA Subsample 665 710-671.26-657.08-510.09 [432.99] [409.15] [319.93] Notes: Bas adjusted robust standard errors are n brackets. * and ** ndcate sgnfcance at 5% and 1% level two tal t-test. For the frst sample, the treatment group s the cohort 1969 and the comparson group s the cohort 1968. For the second sample, the treatment group s cohorts 1969&1970 and the comparson group s cohorts 1967&1968. The pre-treatment perod s 2001 and post-treatment perod s 2002. 34
Table 5: Placebo Law I: Comparng Changes n Total Medcal Expendtures for Cohorts Both Treated or Not Treated Dfference-n-Dfference Covarate Matchng # of Neghbors Matched 1 5 10 Panel A: Cohort 1969 vs. Cohort 1970 MSA Treatment Group Y2002-18.78 MSA Subsample -41.33-25.45-24.09 [24.53] [30.9] [26.41] [25.45] Deductble Treatment Group Y2002-55.51 Deductble Subsample -61.24-71.11-58.52 [48.4] [68.59] [52.65] [52.04] SPA Treatment Group Y2002-517.81 SPA Subsample -880.01-682.04-631.77 [423.94] [473.47] [401.31] [401.82] Panel B: Cohorts 1967 vs. Cohort 1968 MSA Treatment Group Y2002-46.58 MSA Subsample -50.79-48 -45.47 [27.24] [32.1] [28.64] [25.84] Deductble Treatment Group Y2002 79.60 Deductble Subsample 63.81 49.83 59.27 [74.42] [108.92] [93.87] [92.45] SPA Treatment Group Y2002 406.85 SPA Subsample 345.62 800.47 441.32 [369.51] [804.24] [556.31] [446.73] Notes: Robust standard errors are n brackets for the Dfference-n-Dfference model. Bas adjusted robust standard errors are n brackets for matchng. For panel A, both cohorts suffer from a reducton of 1% o f ther annual salares n 2002. We randomly assgn the treatment status to the cohort 1969. For panel B, nether cohort s affected by the polcy change. We randomly assgn the treatment status to the cohort 1968. The pre-treatment perod s 2001 and post-treatment perod s 2002. 35
Table 6: Placebo Law II: Comparng Changes n Total Medcal Expendtures between 2000 and 2001 Dfference-n-Dfference Covarate Matchng # of Neghbors Matched 1 5 10 Panel A: Cohort 1968 vs. Cohort 1969 MSA Treatment Group Y2002 2.08 MSA Subsample 10.69-6.73-17.14 [29.63] [51.58] [35.92] [35.6] Deductble Treatment Group Y2002-109.55 Deductble Subsample 171.08 114.18 127.75 [96.00] [96.62] [90.94] [89.91] SPA Treatment Group Y2002-516.74 SPA Subsample 658.39 425.51 348.04 [334.92] [544.83] [517.47] [402.72] Panel B: Cohort 1967 vs. Cohort 1968 MSA Treatment Group Y2002 7.23 MSA Subsample 5.96 13.93 4.47 [23.03] [30.66] [27.1] [26.31] Deductble Treatment Group Y2002 7.87 Deductble Subsample 44.52 42.6 25.38 [56.52] [61.78] [54.74] [54.23] SPA Treatment Group Y2002 53.47 SPA Subsample -67.03-244.32-184.85 [329.32] [389.32] [334.7] [317.45] Panel C: Cohorts 1969 vs. Cohort 1970 MSA Treatment Group Y2002 0.16 MSA Subsample -2.72-20.4-11.42 [34.18] [49.37] [39.11] [38.87] Deductble Treatment Group Y2002-90.97 Deductble Subsample -156.36-179.06-158.1 [101.19] [224.36] [187.94] [160.56] SPA Treatment Group Y2002 0.45 SPA Subsample 224.51-22.15 97.01 [333.52] [353.57] [354.26] [352.61] 36
Notes: Robust standard errors are n brackets for the Dfference-n-Dfference model. Bas adjusted robust standard errors are n brackets for matchng. For ths placebo law, the pre-treatment perod s 2000 and post-treatment perod s 2001. In other words, we estmate the Dfference-n-Dfference model and matchng as f the change s made n 2001. For panel A, the treatment group s the cohort 1969 and the comparson group s the cohort 1968. For panel B, the treatment group s the cohort 1968 and the comparson group s the cohort 1967. For panel C, the treatment group s the cohort 1970 and the comparson group s the cohort 1969. 37
Table 7: Decomposton of The Effects of a Shock n MSA on Total Medcal Expendtures (Dfference-n-Dfference) (1) (2) (3) Cohort 1969 vs. Cohort 1968 Cohort 1969 vs. Cohort 1970 Cohort 1967 vs. Cohort 1968 MSA Deductble SPA MSA Deductble SPA MSA Deductble SPA Payment Payment Payment Payment Payment Payment Payment Payment Payment Panel A Treatment Group Y2002-11.01* -19.37** -76.78* -2.36-1.96-70.50-1.99 1.49 33.61 [4.48] [5.16] [37.85] [4.30] [5.30] [39.80] [4.96] [5.78] [41.13] Panel B MSA Treatment Group Y2002-13.32* -8.40* 17.37-3.42-3.38-11.98-4.36-6.30-35.93 [6.54] [4.18] [17.60] [6.37] [4.80] [19.79] [7.29] [4.50] [22.38] Deductble Treatment Group Y2002-8.90-54.23** -122.32* -2.38-5.13-48.00-3.36 4.48 78.47 [5.05] [10.60] [58.55] [4.86] [9.81] [42.57] [5.87] [10.70] [68.04] SPA Treatment Group Y2002-4.32-1.57-656.59-8.83 13.48-522.46 16.77 34.33 355.75 [8.06] [33.76] [361.03] [7.65] [34.86] [408.27] [10.78] [38.94] [354.08] # of Enrollees 8269 8269 8269 7786 7786 7786 7773 7773 7773 Notes: Robust standard errors are n brackets for Dfference-n-Dfference. Results usng equaton (9) are presented n panel A. Results usng equaton (10) are presented n panel B. For the frst sample, the treatment group s the cohort 1969 and the comparson group s the cohort 1968. For the second sample, the treatment group s the cohort 1969 and the comparson group s the cohort 1970. For the thrd sample, the treatment group s the cohort 1968 and the comparson group s the cohort 1967. 38
Table 8: Estmates of Substtuton Effects (Dfference-n-Dfference: 2001 vs. 2003) (1) (2) Cohort 1969 vs. Cohort 1968 Cohorts 1969&1970 vs. Cohorts 1967&1968 Panel A Treatment Group Y2002-38.18-2.25 [42.15] [33.79] Panel B MSA Treatment Group Y2002 23.94 26.62 [25.33] [21.18] Deductble Treatment Group Y2002 27.95 52.30 [53.74] [55.97] SPA Treatment Group Y2002-814.02* -498.68 [405.52] [294.74] Panel C Lower Quntle Treatment Group Y2002 35.92 22.11 [41.17] [29.45] Mddle Quntle Treatment Group Y2002-15.69 2.51 [39.97] [41.34] Upper Quntle Treatment Group Y2002-177.12-68.34 [112.47] [86.24] # of Enrollees 7884 14802 Notes: Bas adjusted robust standard errors are n brackets. * and ** ndcate sgnfcance at 5% and 1% level two tal t-test. For the frst sample, the treatment group s the cohort 1969 and the comparson group s the cohort 1968. For the second sample, the treatment group s cohorts 1969&1970 and the comparson group s cohorts 1967&1968. The pre-treatment perod s 2001. The post-treatment perod s 2003 rather than 2002. We nvestgate whether enrollees substtute medcal expendtures n 2002 wth those n 2003. 39
Fgure 1. Zhenjang Health Care Fnancng System Contrbuton Dstrbuton Payment Structure Employer Contrbuton 8% of Annual Salary Socal Poolng Account 6% Socal Poolng Account Indvdual Contrbuton 2% of Annual Salary Medcal Savngs Account 4% 10% of Annual Salary Deductble Medcal Savngs Account Note: Ths graph shows the health nsurance for enrollees younger than 44 years old n Zhejang n 2001.
Fgure 2. Enrollees Dscount the MSA Money (e.g. 100%) "True" Cost of Medcal Expendture B Before Polcy Change C After Polcy Change B C A A O 300 400 1300 1400 Medcal Expendture 41
Fgure 3. Enrollees Treat the MSA Money the Same as Out-of-pocket Money C "True" Cost of Medcal Expendture A 45 Degree Lne A B B Before Polcy Change C After Polcy Change O 300 400 1300 1400 Medcal Expendture 42
Fgure 4. MSA Contrbuton, MSA Balance and Total Medcal Expendture (2000-2003) 900 RMB (Chnese Currency) 800 700 600 500 400 300 2000 2001 2002 2003 Total Medcal Expendture(69) Balance of MSA(69) Contrbuton to MSA(69) Total Medcal Expendture(68) Balance of MSA(68) Contrbuton to MSA(68) 43