Work, Offers, and Take-Up: Decomposing the Source of Recent Declines in Employer- Sponsored Insurance

Work, Offers, and Take-Up: Decomposng the Source of Recent Declnes n Employer- Sponsored Insurance Lnda J. Blumberg and John Holahan The Natonal Bureau of Economc Research (NBER) determned that a recesson began n March 2001 (NBER 2001). And whle NBER also determned that the recesson offcally ended n November of 2001, the economy contnued to be sluggsh well past that tme, partcularly n employment. Evdence showed that the rate of nsurance coverage also decreased durng that perod, prmarly because of a drop n the rate of employer-sponsored nsurance (ESI) coverage (Holahan and Wang 2004). Prevous research has not explored, however, the factors underlyng the declne n ESI that occurred durng ths economc downturn. Although the employment rate clearly declned for the populaton as a whole durng ths tme perod, the probablty of beng offered health nsurance and the probablty of takng up the coverage offered are also factors affectng the lkelhood of ESI coverage. Ths analyss assesses how these components of coverage changed durng the economc downturn. We examne how these changes vared for unmarred adults, marred adults, and dependent chldren, and assess how much of the declne n coverage that concded wth the downturn was attrbutable to changes n employment rates versus changes n offer rates versus changes n take-up rates. We also determne how the changes over ths perod dffered for low-ncome and hgher-ncome populatons, and how changes n the components of coverage vared by employer sze. The next secton descrbes the data used for ths study and defnes the key analytc varables. The thrd secton provdes a detaled explanaton of the methodologcal approach taken. The fourth secton presents the emprcal results, and the ffth provdes a summary of the results and conclusons of the analyss. The key results from ths analyss are summarzed n the table and bullets below. Change n ESI coverage Summary of Man Results (percent) Share due to Share due to changes n changes n work offer rates Share due to changes n take-up Share due to other factors All -2.2% ** 27.8% 7.2% 64.4% 0.6% < 200% of FPL -4.6% ** 26.1% 24.2% 48.1% 1.7% > 200% of FPL -1.3% -8.9% 12.2% 91.0% 5.6% 1. There was a sharp drop n ESI coverage between 1999 and 2002. In 1999, 69.2 percent of the nonelderly populaton had employer-sponsored nsurance coverage. In 2002, the ESI coverage rate had fallen to 66.9 percent, a change of 2.2 percentage ponts. Had the rate of ESI coverage stayed at 1999 levels, 5.9 mllon more ndvduals would have had coverage n 2002 than was actually the case. 1

2. All three groups dependent chldren, unmarred adults and marred adults experenced declnes n ESI. Dependent chldren had the largest declnes (3.1 percentage ponts), followed by unmarred adults (2.3 percentage ponts) and marred adults (1.7 percentage ponts). 3. Almost two-thrds of the declne n ESI for the overall populaton was attrbutable to the decrease n the lkelhood of takng up an employer offer. Take-up rates declned by 1.7 percentage ponts for unmarred adults, 1.0 percentage pont for marred adults, and 3.5 percentage ponts for dependent chldren. 4. Most of the remanng loss of ESI was the result of a reducton n the lkelhood of work, whch accounted for almost 30 percent of the declne. The probablty of not workng ncreased by 1.9 percentage ponts for unmarred adults. The probablty that a chld had no workng parent ncreased by 1 percentage pont. 5. Declnes n ESI coverage were partcularly sharp for the low-ncome populaton. ESI coverage for the low-ncome nonelderly populaton fell by just under 5 percentage ponts. Ths amounts to a declne of about 12 percent, substantally larger than the drop (less than 2 percent) experenced by the hgher-ncome nonelderly populaton. 6. The low-ncome populaton was affected by a reduced lkelhood of work, fewer offers, and lower take-up rates. For the low-ncome populaton as a whole, the declne n take-up accounted for about 48 percent of the decrease n the rate of ESI. The drop n work accounted for about 26 percent of the change and reduced offer rates for about 24 percent. 7. The share of low-ncome people wth at least one adult n the mmedate famly employed by an establshment fell by 3.1 percentage ponts. Low-ncome unmarred adults experenced the largest declne n establshment-based work (3.7 percentage ponts, compared wth roughly 2.5 percentage ponts for marred adults and chldren). 8. The share of low-ncome persons wth at least one employer offer n the famly fell by 2.7 percentage ponts. The largest declne n offer was experenced by marred adults (4.8 percentage ponts). The declnes n take-up rates for low-ncome Amercans were even more dramatc, fallng by 5.8 percentage ponts. Take-up rates by unmarred adults fell by 4.7 percentage ponts, and for chldren by 8.9 percentage ponts. 9. The declne n ESI coverage for hgher-ncome ndvduals fell by just over 1 percentage pont between 1999 and 2002. Only take-up declned sgnfcantly for ths ncome group overall, droppng by 1.5 percentage ponts. Unmarred adults n ths ncome group experenced a declne n offers, whereas marred adult and chldren s take-up rates declned. 10. There was a shft n work from large to small employers, where the lkelhood of ESI coverage s lower. The probablty of workng for a small employer ncreased and the probablty of workng for a large employer declned for unmarred adults. There was also a modest ncrease n the probablty of workng n government for ths group. At the same tme, offer rates declned n small establshments. Thus, whle employment of unmarred adults was shftng to small establshments and government, the offer rates for those workers were fallng. For those workng n small establshments, there were also declnes n take-up rates of over 5 percentage ponts. These results suggest that ncreased unemployment, reduced offers of coverage, and declnng take-up rates all contrbuted to the declne n employer-sponsored nsurance. Declnes 2

n employment wll drectly affect the share of the populaton wth employer-sponsored nsurance. Declnes n employer offers occur prmarly wthn small frms, but also reflect shfts from full-tme to part-tme work and from employment n large establshments to small establshments. The declnes n take-up are consstent wth the reductons n employer contrbutons to employee health nsurance premums shown n the Natonal Survey of Amerca s Famles and surveys of health benefts sponsored by the Kaser Famly Foundaton (Holahan 2003; Gabel et al. 2003). Declnes n take-up rates are also consstent wth earler studes that have shown declnes n take-up of employer offers from the late 1980s to the late 1990s (Farber and Levy 2000; Cooper and Schone 1997). Data and Varable Defnton Ths analyss uses the 1999 and 2002 Natonal Surveys of Amerca s Famles (NSAF). The NSAF s a household survey that provdes nformaton on more than 100,000 chldren and nonelderly adults representng the nonnsttutonalzed Unted States populaton under age 65. Low-ncome populatons n 13 states are oversampled. Detaled nformaton s collected for adults and up to two chldren n each famly. The survey ncludes nformaton on employment status, health nsurance coverage, employer sponsorshp of health nsurance plans, and ncome. Detaled nformaton on demographc characterstcs, access to health care servces, and famly well-beng are also collected. Issues Related to Varable Defnton Two ssues related to defnng varables deserve dscusson n some detal. The frst s the nature of the NSAF varable ndcatng the presence of an employer offer, whch s dfferent from offer varables n other natonally representatve data sets. The second s the computaton of the ncome varable used to determne ncome relatve to poverty. Because the NSAF provdes total ncome for the year before the survey (1998 and 2001), but the offer nformaton relates to the job held at the tme of the ntervew (1999 or 2002), we mputed survey year ncome to each ndvdual n the analyss. Employer Offer. The NSAF queston about the exstence of an employer offer reads, Does your current employer offer health nsurance to workers n the same poston as yours? It does not ask whether the respondents themselves are drectly offered employer nsurance. The queston was worded n a way to elct nformaton about the type of job that the person held, not about the person s own experence. Consequently, ndvduals can answer yes to the queston even f they are not elgble to enroll n an employer plan. It s mportant to keep ths n mnd, because offer rates produced usng the NSAF wll be hgher than those based on other data sets that focus on whether a partcular ndvdual receved an offer. Addtonally, offer rates based on the NSAF wll be lower than those computed from surveys that ask f an ndvdual s employer offers health nsurance to any workers at the frm where the ndvdual s employed. Most mportant for these purposes, however, the offer queston was asked consstently n the 1999 and 2002 surveys. Consequently, the queston should accurately reflect the change n offer probabltes over tme. Decomposng the source of the change n employer-based coverage over ths perod s precsely the goal of ths analyss, makng the use of the NSAF offer varable n ths context approprate. It should be noted, however, that a declne n the NSAF data n the probablty of takng up an offer could reflect an ncrease n the rate of worker declnes, an ncrease n the lkelhood that employers are offerng coverage to some workers but not others holdng smlar postons, or a combnaton of both. Unfortunately, no data can be used to assess whether a change n the probablty of elgblty has occurred durng ths perod. Although the 3

Current Populaton Survey was conductng an employee benefts supplement to ther February survey n odd-numbered years that could have answered ths queston, the supplement has been dscontnued. The last avalable data are from 2001. Income. Ideally, the measure of ncome used n ths analyss would correspond to the perod referenced n the employer offer and coverage questons. Respondents are asked about current offer and coverage status; however, total ncome s computed based on the prevous calendar year. The NSAF does nclude nformaton on current (.e., tme-of-survey) earnngs and current number of hours a week worked. We use ths data, along wth that on unearned ncome n the year before the survey to mpute survey-year total ncome for each ndvdual. Ths mputed ncome varable s then used to categorze ndvduals by ncome relatve to poverty. We use the current earnngs nformaton to construct earnngs per hour, and multply ths value by the number of hours worked per week. For those ndvduals who have not changed jobs n the last 12 months, we assume that ther current weeks worked per year s the same as that reported for the prevous year. For those who have changed jobs, we mpute number of weeks worked per year based on those ndvduals who dd not change jobs. We use a regresson-based statstcal matchng approach for ths mputaton. Number of weeks per year s estmated on the sample of ndvduals who have not changed jobs n the last 12 months ( donors ), usng ndependent varables for self-employment status, gender, hourly earnngs, occupaton, ndustry, and an nteracton between occupaton and ndustry. The results of ths regresson are used to predct weeks per year for those wth recent job changes or those startng employment after a perod wthout work ( mputees ) and for the donors. The predcted values of the mputees are compared to the predcted values for the donors. Each mputee s assgned the actual number of weeks worked n the last year from the donor whose predcted value s closest to the mputee s own predcted value. Current earnngs per week are then multpled by the number of weeks per year for each worker to compute current-year annual earnngs. Unearned ncome for the current year s assumed to be largely consstent wth past-year unearned ncome. The exceptons are that we do not carry forward those components of unearned ncome that were ether unlkely to be avalable n the followng year because of tme lmts or are by ther nature unstable (unemployment compensaton, emergency/one-tme cash payments from a welfare program, vouchers or coupons to help pay for specal expenses, fnancal support from frends/relatves outsde household) or are assocated wth havng extremely low ncome (general assstance, TANF/AFDC, food stamps). In addton, to take nto account the volatlty of the stock market durng ths perod, we adjusted nterest from sources such as bank accounts, money markets or certfcates of deposts, dvdends from stocks, or mutual funds for average changes between 1998 and 1999 and between 2001 and 2002. We computed the average value of unearned ncome from these sources for those wth some ncome usng the 1999, 2000, 2002, and 2003 fles of the March Current Populaton Survey (CPS). These survey years represent data years 1998, 1999, 2001, and 2002, respectvely. We appled the percentage change n the means between the 1999 and 2000 CPS surveys (9.35 percent) to the unearned ncome of ths type for ndvduals n the 1999 NSAF, and the change between the 2002 and 2003 CPS surveys (-23.30 percent) to the unearned ncome of ths type for ndvduals n the 2002 NSAF. Total ncome n the survey year for each ndvdual was then computed by addng the mputed values of survey year earned and unearned ncome. Famly ncome was computed by summng total survey-year ncome over members of the same famly. 4

Methodologcal Approach Ths analyss examnes the components of the change n ESI between 1999 and 2002. In a smple world, an ndvdual s probablty of havng ESI can be defned as ESI ˆ = Wˆ * Oˆ * Tˆ, where Wˆ s the probablty of beng a worker, Ô s the probablty of havng an employer offer of health nsurance, condtonal on beng a worker, and Tˆ s the probablty of takng up an employer offer, condtonal on recevng such an offer. Ths framework s complcated, however, by marred adults and dependent chldren. Marred adults may obtan ESI through ther own place of employment or through ther spouse s employment, and dependent chldren s access to ESI s through a parent. In addton, for each subpopulaton (unmarred adults, marred adults, and dependent chldren), some ndvduals report havng ESI coverage but do not have a current offer of coverage, ether through ther own or an observable famly member s place of employment. These ndvduals may have coverage through a former employer or through a famly member not present n the household. We address these complcatons n the followng ways. Frst, each of the three subpopulaton groups s treated separately. The rate of ESI coverage for unmarred people s defned as ESI ˆ unmarred = ( Wˆ unmarred * Oˆ unmarred * Tˆ unmarred ) + Aˆ unmarred, where W, O, and T are computed for the unmarred adult populaton, and A unmarred s the share of unmarred adults reportng alternatve sources of ESI coverage (e.g., through a former employer, through a famly member not present n the household). marred For marred adults, we compute: ESI ˆ = ( Wˆ * Oˆ * Tˆ ) + Aˆ marred marred marred marred where W marred, refers to the probablty that at least one spouse s a worker; O marred s the probablty that at least one of the employed spouses works for an offerng employer, and T marred s the probablty of the worker takng up coverage through ther own or a spouse s employer, gven that at least one of them works for an offerng establshment. chld For dependent chldren, the probablty of ESI coverage s computed as: ESI ˆ = ( Wˆ * Oˆ * Tˆ ) + Aˆ, parent parent chld chld where Wˆ parent s equal to the probablty that the chld has at least one workng parent; Ô parent s equal to the probablty that at least one parent (gven s/he works) s offered health nsurance, and Tˆ chld s the probablty that the chld has ESI coverage through a parent wth an offer. For purposes of ths analyss, work (W) s defned as beng employed n some type of establshment, and where employment s compensated wth cash wages. Self-employment s not ncluded n W, and s grouped together wth nonworkers and those workng wthout pay n A, snce these ndvduals cannot receve offers of employer-sponsored nsurance. The change n the probablty of nonwork, self-employment, and nformal work over ths perod are examned separately. For each subpopulaton (unmarred adults, marred adults, and dependent chldren), the share workng, the share offered ESI (gven work), the share takng up ESI (gven an offer), and 5

the share wth alternatve sources of ESI were computed n 1999 and 2002. In addton to the overall changes n each component of ESI coverage, we decomposed the change n ESI for each subpopulaton and for the populaton as a whole nto ts components. The decomposton was computed for each subpopulaton n the followng way. 1. A hypothetcal probablty of ESI coverage n 2002 was computed, holdng the probablty of E SI ˆ hyp1, = Wˆ 1999, * Oˆ 2002, * Tˆ 2002, + Aˆ 2002, work at the 1999 level ; 2. A hypothetcal probablty of ESI coverage n 2002 was computed, holdng the probablty of E SI ˆ hyp2, = Wˆ 2002, * Oˆ 1999, * Tˆ 2002, + Aˆ 2002, offer at the 1999 level ; 3. A hypothetcal probablty of ESI coverage n 2002 was computed, holdng the probablty of E SI ˆ hyp3, = Wˆ 2002, * Oˆ 2002, * Tˆ 1999, + Aˆ 2002, take-up at the 1999 level ; 4. A hypothetcal probablty of ESI coverage n 2002 was computed, holdng the probablty of E SI ˆ hyp4, = Wˆ 2002, * Oˆ 2002, * Tˆ 2002, + Aˆ 1999, alternatve ESI coverage at the 1999 level Delta hyp1 = ESI ˆ hyp1, ESˆ I1999, Delta hyp2 = ESI ˆ hyp2, ESˆ I1999, 5. ; ; Delta hyp3 = ESI ˆ hyp3, ESˆ I1999, Delta hyp4 = ESI ˆ hyp4, ESˆ I1999, ; and were each computed for each subpopulaton. These fgures represent the rate of ESI coverage for a gven subpopulaton, f a partcular component (W, O, T, A) of coverage had not changed from ts 1999 level. 6. Each value of Delta hyp was then subtracted from the actual change n ESI over the perod (ESI 2002, ESI 1999, ), to compute the porton of the actual change attrbutable to each component (W, O, T, A) of nsurance coverage. 1 7. The amount of the change attrbutable to each component was then dvded by the total change n ESI for the applcable subpopulaton over the perod, to compute the percentage of the change attrbutable to each component. 8. The share of the change n the rate of ESI coverage for the whole populaton attrbutable to each component was computed usng the calculatons descrbed above, as well as the share of the populaton attrbutable to marred adults, unmarred adults, and dependent chldren. For ths computaton, the amount of the ESI change attrbutable to change n the dstrbuton of the 3 subpopulaton groups s also ncluded. In addton to computng the overall changes, changes n the components of ESI coverage are also broken down by frm sze and wthn ncome relatve to poverty groups. The change n ESI coverage wthn each ncome group was also decomposed to demonstrate the relatve mportance of each component of the change. 1 The sum of the portons of the ESI change attrbutable to each component do not add up precsely to the actual overall change n each subpopulaton because of small nteractve effects of the components. However, these nteractve effects are very small (n all cases less than one-tenth of a percentage pont); consequently, the nteractons are gnored, and the dfferences are normalzed away. 6

Results Changes n Work, Offer, and Take-Up for the Nonelderly In 1999, 69.2 percent of the nonelderly populaton had employer-sponsored nsurance coverage. In 2002, the ESI coverage rate had fallen to 66.9 percent, a change of 2.2 percentage ponts, and a relatve declne of over 3 percent. Had the rate of ESI coverage stayed at 1999 levels, 5.9 mllon more ndvduals would have had coverage n 2002 than was actually the case. Table 1 shows how each component of ESI coverage (the probablty of work, the probablty of havng access to an offer gven a workng famly member, 2 and the probablty of takng up coverage gven an offer) changed between 1999 and 2002 for each subpopulaton (unmarred adults, marred adults, and dependent chldren). 3 The table also shows how the probablty that an ndvdual has ESI through an alternatve source has changed over tme as well as how each subpopulaton has changed as a share of the entre populaton. Lookng at the percentage pont changes n the bottom secton of the table, dependent chldren experenced the largest declne n ESI durng ths perod, 3.1 percentage ponts. Unmarred adult ESI declned by 2.3 percentage ponts, and marred adult coverage fell by 1.7 percentage ponts. Each change was statstcally sgnfcant. Across the full populaton, only the probablty of takng up ESI (condtonal on the exstence of an offer) changed sgnfcantly, declnng by 1.9 percentage ponts. Consstently, for each subpopulaton, the component wth the largest percentage pont declne, and the only one that s statstcally sgnfcant, s take-up. Take-up declned by 1.7 percentage ponts for unmarred adults, 1.0 percentage pont for marred adults, and 3.5 percentage ponts for dependent chldren. Because of the specfc offer queston ncluded n the NSAF and descrbed above, the declne n take-up calculated here could be attrbutable, at least n part, to a change n the lkelhood of elgblty for an employer offer. We presume that the vast majorty of the change n take-up was actually attrbutable to an ncreased probablty that workers were actually declnng offers, probably because of ncreased employee contrbutons (Holahan 2003), but we cannot test ths assumpton emprcally for ths tme perod. 4 For the full populaton, the probablty of havng a workng adult n the mmedate famly declned by.7 percentage ponts, but that change was not sgnfcant. Unmarred adults saw a decrease n the lkelhood of employment of 1.4 percentage ponts, also a statstcally nsgnfcant change. Unmarred adults saw larger declnes than marred adults n the lkelhood of recevng an offer gven employment (1.5 percentage ponts compared wth.1 percentage ponts for marred adults), although ths declne was not statstcally sgnfcant ether. The probablty of havng ESI through an alternatve source dd not change apprecably over the perod. The subpopulaton dstrbuton was also qute stable over the perod, however; the.4 percentage pont declne for chldren as a share of the nonelderly populaton was statstcally sgnfcant. 2 Famly, as used here, refers to members of the same health nsurance unt. Health nsurance unts nclude those ndvduals n a famly who typcally can purchase nsurance under a sngle polcy marred adults, ther chldren up to age 18, and chldren younger than 23 who are full-tme students. 3 For purposes of ths analyss, beng marred s defned as havng a spouse present n the household. Consequently, the unmarred category n these tables ncludes some marred adults who are separated or otherwse have a spouse not currently resdng n the home. 4 It s worth notng, however, that author tabulatons of the 1999 and 2001 Contngent Worker Supplements to the Current Populaton Survey showed that at least between 1999 and 2001, the rate of elgblty gven that an employer sponsored an nsurance plan dd not declne. 7

Decomposton of the Sources of Change n ESI Coverage Table 2 shows the share of the change n ESI attrbutable to each component for each subpopulaton and the nonelderly populaton as a whole. Row 1 shows that the declne n coverage for unmarred adults was roughly evenly splt between work, offer, and take-up. Coverage through alternatve sources actually ncreased slghtly over the perod studed, very modestly offsettng some of the declne, hence the negatve value n column A of the table. The majorty of the change (69 percent) n ESI for marred adults s attrbutable to a declne n take-up. A declne n the probablty of work was less than half as mportant as the decrease n take-up, and a declne n the probablty of havng an alternatve source of ESI coverage was also of less mportance, but stll notable. The probablty of offer (through at least one spouse) actually ncreased for marred people, offsettng the declne for ths group to a modest extent. A declne n the probablty of take-up was also the most mportant factor for the declne n ESI coverage among chldren, accountng for 74 percent of the declne. The second most mportant change for ths group was the declne n the probablty of work for at least one parent, whch accounted for about 18 percent of the drop n ESI. The probabltes of offer and alternatve ESI coverage also fell, but were of relatvely lttle mportance to the overall declne n ESI for chldren. Taken together, almost two-thrds of the declne n ESI for the nonelderly populaton was attrbutable to the decrease n the lkelhood of takng up an employer offer, wth the declne n the probablty of work accountng for less than 30 percent of the declne. The modest shft of the populaton dstrbuton actually offset a slght amount of the declne. The Declne n Establshment-Based Work Table 3 provdes detaled nformaton on the changes underlyng the declne n work wthn establshments. As shown n table 1, the probablty of work n establshments declned between 1999 and 2002. We separated ths declne n establshment work nto ts component parts: nonwork, self-employment, and other work (defned as temporary work or work wthout pay). As table 3 shows, there was a sgnfcant ncrease n nonwork among unmarred adults and marred adults durng ths perod and a modest (statstcally nsgnfcant) declne n selfemployment. Whle there was also some change n the other work category, the category s persstently very small, so we do not dscuss t here. The probablty of not workng ncreased by 1.9 percentage ponts for unmarred adults and by 1.0 percentage pont for marred adults between 1999 and 2002. 5 The probablty that a chld had no workng parent grew as well, by 1 percentage pont, although ths change was not statstcally sgnfcant. Whle some have assumed that self-employment would have grown n the perod, somewhat compensatng for the loss of employment-based labor market durng the recesson, these data do not bear that assumpton out. Self-employment remaned essentally constant wthn each subpopulaton. 5 To be precse, ths 1 percentage pont ncrease reflects the change n probablty that a marred person does not work and does not have a workng spouse. 8

Change n Work, Offer, and Take-Up for the Low-Income Populaton Table 4 shows the changes n the components of ESI coverage for the low-ncome populaton, defned as those n famles wth ncomes below 200 percent of the federal poverty level (FPL) n each year. In 1999, the FPL for a famly of four resdng n the contnental Untes States was $16,895; n 2002, t was $18,244. 6 As row 4 of the bottom secton of the table shows, ESI coverage for the low-ncome nonelderly populaton fell by 4.6 percentage ponts. Ths amounts to a relatve declne of about 12 percent. Ths drop s substantally larger than that experenced by the hgher-ncome populaton, who saw ESI coverage fall by 1.3 percentage ponts (not sgnfcant), a relatve declne of under 2 percent (fgures for the hgher-ncome populaton are shown n table 7, and wll be dscussed further below). Vrtually every component of coverage declned between 1999 and 2002 for the lowncome populaton. The probablty of havng an adult n the mmedate famly employed by a frm fell by 3.1 percentage ponts, the probablty that at least one of the workng adults n the famly was employed by an offerng frm fell by 2.7 percentage ponts, and the probablty of takng up employer coverage gven an offerng employer fell dramatcally, by 5.8 percentage ponts. The share of unmarred low-ncome adults workng n establshments fell by 3.7 percentage ponts, whle the share of marred adults where at least one spouse was workng fell by 2.5 percentage ponts (not sgnfcant). The probablty that a chld had at least one workng parent declned by 2.3 percentage ponts. The offer rate declned sgnfcantly for low-ncome marred adults (4.8 percentage ponts) and for low-ncome chldren (2.8 percentage ponts). Take-up rate declnes were qute dramatc for unmarred adults and dependent chldren. Take-up among unmarred adults fell by 4.7 percentage ponts and take-up for chldren dropped by 8.9 percentage ponts. (The large declne n ESI for chldren n ths ncome group could reflect publc expansons, but ths would not explan the drop for unmarred adults.) Taken together wth changes n the share wth coverage from alternatve sources, ESI coverage rates of low-ncome unmarred adults fell by 1.5 percentage ponts, and ESI coverage rates of low-ncome marred adults and chldren fell by 6.6 and 5.5 percentage ponts, respectvely. These results mean that 3.6 mllon low-ncome adults and chldren dd not have ESI n 2002 that would have had t f coverage rates had remaned at 1999 levels. Decomposton of the Sources of Change n ESI Coverage for the Low-Income Populaton Table 5 shows how the declne n ESI coverage for the low-ncome populaton can be attrbuted to each component of coverage. Row 1 of the table shows that the declne n work and take-up are the most mportant factors accountng for the declne n ESI for low-ncome unmarred adults. Offer stayed vrtually constant for ths group, and alternatve sources of coverage ncreased, accountng for the negatve value n column A. For marred low-ncome adults, 37 percent of the declne s attrbutable to the decreased probablty of offers, 20 percent s attrbutable to the declne n work, and about 17 percent to the declne n take-up. Interestng for ths group s the mportance of the declne n ESI coverage through alternate sources, whch accounts for 27 percent. Gven that both spouses are present for ths subpopulaton, ths change lkely sgnfes decreased coverage through COBRA durng ths perod. 6 Income relatve to poverty vares by the number of famly members. We provde the FPL for a famly of four for llustratve purposes. 9

The declne n ESI for low-ncome dependent chldren, however, s most attrbutable to decreased take-up of parental offers. The declne n the probablty of havng a workng parent and n the probablty of offer were about equally mportant as contrbutors to the declne for chldren. There was a small ncrease n the probablty of chldren havng an alternatve source of ESI, whch slghtly offset the declnes n the other components. Takng all three subpopulatons nto consderaton, the declne n take-up accounted for about 48 percent of the decrease n the rate of ESI, the drop n work accounted for 26 percent of the change, and decreased offer accounted for 24 percent. Table 6 provdes detals on the declne n the probablty of workng n an establshment settng for the low-ncome populaton. The ncrease n the probablty of not workng about 3 4 percentage ponts for each subpopulaton s solely responsble for the declne n establshmentbased work. Self-employment dd not change sgnfcantly for ths populaton. Change n Work, Offer, and Take-Up for the Hgher-Income Populaton Table 7 provdes detaled nformaton on the changes n the components of ESI coverage for those ndvduals wth ncomes at or above 200 percent of FPL. As already noted, the declne n ESI coverage for ths hgher-ncome group was much smaller over ths perod than was the case for the low-ncome group. Ther probablty of coverage fell by 1.3 percentage ponts (not sgnfcant) between 1999 and 2002. Ths mples that 2.4 mllon hgher-ncome people would have had ESI n 2002 f the ESI coverage rates had remaned at 1999 levels. Across all subgroups of the hgher-ncome populaton, only take-up declned sgnfcantly, by 1.5 percentage ponts. There was no sgnfcant change among the hgher-ncome group n the probablty of work. Offer rates fell by 2.3 percentage ponts for hgher-ncome unmarred adults and stayed vrtually constant for marred adults and dependent chldren. The probablty of takng up coverage (gven an offer) fell for dependent chldren and marred adults. Overall, dependent chldren n hgher-ncome famles saw the largest declne n ESI, 2.5 percentage ponts. Adult ESI coverage n ths ncome group fell, but the change was not statstcally sgnfcant for the unmarred or the marred subpopulatons. Decomposton of the Sources of Change n ESI Coverage for the Hgher-Income Populaton Table 8 provdes the results of the decomposton of the declne n ESI coverage for the hgherncome populaton as a whole and separately for unmarred adults, marred adults, and dependent chldren. The declne n coverage for unmarred adults was attrbutable n large part to lower rates of employer offers. The declnes n take-up were also mportant, but they accounted for roughly a thrd as much of the declne as dd offer. The declnes n coverage for marred adults and for dependent chldren, however, were clearly domnated by declnes n take-up. Both marred adults and chldren experenced a decrease n the probablty of recevng ESI coverage through an alternatve source, although ths factor was of modest mportance compared wth the declne n take-up. Across all three subpopulatons, decreased take-up accounted for the vast majorty of the coverage declne among the hgher-ncome populaton. Because the probablty of workng n an establshment dd not declne durng ths perod for the hgher-ncome populaton, we do not present a separate table decomposng that change, as we dd wth the low-ncome populaton. 10

Changes n Work, Offer, and Take-Up by Establshment Sze Table 9 allows us to examne how the changes n the components of ESI vared by sze of employer. For these purposes, small s defned as an establshment of fewer than 25 workers. Informaton on government workers s kept separate from the small and large employers because the NSAF does not ask government workers for the sze of ther establshment. Over the course of ths perod, the probablty of workng for a small employer ncreased and the probablty of workng for a large employer decreased. These changes were statstcally sgnfcant for unmarred adults and, wth regard to the decrease n large employer work, for the parents of dependent chldren. 7 The most notable change was a decrease of 3.4 percentage ponts n the probablty of an unmarred adult workng for a large employer. There was also a modest ncrease n the probablty of unmarred adults workng n government. The change n the probablty of government work was not sgnfcant for ether marred adults or the parents of dependent chldren. Percentage pont changes n the probablty of offer were greatest for unmarred adults. Whle the probablty of offer decreased by almost 5 percentage ponts for unmarred adults workng n small establshments, and almost 3 percentage ponts for those n government work (although the latter was not statstcally sgnfcant), the probablty of offer for those n large establshments ncreased by over 2 percentage ponts (ths change was not sgnfcant ether). So whle employment of unmarred adults was shftng to small and government employers, the probablty of offer for those workers was fallng. Changes n the probablty of recevng an offer through small establshments also decreased for marred adults and dependent chldren, although the percentage pont declne n offer for those groups was consderably smaller than for unmarred adults and were not statstcally sgnfcant. Ths table also shows that the percentage pont declnes n take-up durng ths perod were greatest for unmarred adults and chldren n small establshments and for marred adults n government. The take-up of unmarred adults employed by small establshments fell by 5 percentage ponts, compared wth only 1 percentage pont for large establshment workers, and vrtually no change for government workers. Take-up of coverage for dependent chldren fell by 7.5 percentage ponts when the only employed parents worked for small employers, compared wth 2 3 percentage pont declnes for those wth parents employed n government or large establshments. Contrary to the pattern exhbted by unmarred adults and chldren, marred adult take-up rates fell by 4.2 percentage ponts n the government and declned only very modestly among small and large employers. These dramatc dfferences n the probablty of takng up over ths perod may reflect larger relatve ncreases n health nsurance premums for small employers between 1999 and 2002. Whle take-up had already been somewhat lower for workers employed by small employers than for those employed by large and government employers, recent changes have clearly exacerbated those dfferences. It s a sgnal that ESI s gettng more dffcult to afford through small employers, and that ther premum ncreases may be relatvely larger than those of large employers. Ths hypothess s also consstent wth the evdence that small employers are 7 Chldren wth at least one parent workng for a government employer were classfed n that group; chldren wth no parent workng for the government, but at least one parent n a large establshment, were classfed as large; and those wth parents only employed n small busnesses were grouped n small. 11

becomng less lkely to offer coverage, and that that declne n offer s greater than s the case for large employers and for government employers. Summary and Conclusons Ths paper has found that the sharp drop n employer-sponsored nsurance between 1999 and 2002 for the nonelderly populaton as a whole was the result of a combnaton of reduced employment levels and lower take-up rates. Overall, about two-thrds of the declne n ESI coverage was attrbutable to lower take-up rates. Ths could reflect changes n the nature of the coverage offered (e.g., ncreased employee contrbutons and fewer benefts) or competng demands on famly budgets. The declnes n rates of employer sponsored nsurance were partcularly sharp for lower-ncome populatons. Ths populaton experenced mportant declnes n all components of ESI coverage: employment, offer, and take-up. Ths s consstent wth the low-ncome populaton beng more greatly affected by the recesson n terms of job loss and premum ncreases. Fnally, there was a shft n work from large to small employers, where the lkelhood of ESI offers s lower. The mportant queston s whether any of these changes are lkely reverse as the economy mproves. Thus far, ncreased rates of economc growth have not led to a sgnfcant ncreases n employment, but ths s lkely to change eventually. To the extent that lower take-up rates are the result of hgher premums and/or hgher employee contrbutons, the lower lkelhood of takng up employers offers may be slower to change because the double-dgt premum nflaton of recent years s unrelated to the economc downturn. To the extent the market for low-wage workers remans depressed because of slow economc growth and global competton, ncreases n overall compensaton are lkely to be slow. If growth n health nsurance premums contnues faster than the growth n wages, take-up rates among low-ncome populatons, partcularly those workng n small frms, could reman depressed and possbly fall even further. References Cooper, P.F., and B.S. Schone. 1997. More Offers, Fewer Takers for Employment-Based Health Insurance: 1987 and 1996. Health Affars 16(6): 142 49. Farber, H.S., and H. Levy. 2000. Recent Trends n Employer-Sponsored Health Insurance Coverage: Are Bad Jobs Gettng Worse? Journal of Health Economcs 19(1): 93 120. Gabel, Jon, Gary Claxton, Ern Holve, Jeremy Pckregn, Hed Whtmore, Kelley Dhont, Samantha Hawkns, and Dane Rowland. 2003. Health Benefts n 2003: Premums Reach Thrteen-Year Hgh as Employers Adopt New Forms of Cost Sharng. Health Affars 22(5): 117 26. Holahan, John. 2003. Changes n Employer-Sponsored Health Insurance Coverage. Snapshots of Amerca s Famles III No. 9. Washngton, DC: The Urban Insttute. Holahan, John, and Mare Wang. 2004. Changes n Health Insurance Coverage durng the Economc Downturn: 2000 2002. Health Affars Web Exclusve W4-31. Natonal Bureau of Economc Research. 2001. The Busness-Cycle Peak of March 2001. Report of the Busness Cycle Datng Commttee. http://www.nber.org/cycles/recessons.html 12

Acknowledgments Ths research was funded by a grant from the Robert Wood Johnson Foundaton. The vews expressed are those of the authors and do not necessarly reflect the postons of the Robert Wood Johnson Foundaton or the Urban Insttute. The authors are grateful to Fredrc Blavn and John Graves for ther research assstance. 13

Table 1. Changes n the Components of Employer-Sponsored Insurance Coverage Rates, 1999-2002 (percent) Survey Year: 1999 S W O T A C Share who Share wth Share who Share wth take up alt. ESI work offer offer source a Share of populaton ESI coverage 1) Unmarred adults 26.8% 70.3% 70.8% 87.9% 13.7% 57.5% 2) Marred adults 41.2% 85.1% 90.0% 92.9% 8.7% 79.9% 3) Dependent chldren 32.0% 76.7% 84.6% 86.5% 9.0% 65.1% 4) Total 100.0% 78.4% 83.7% 89.9% 10.2% 69.2% Survey Year: 2002 1) Unmarred adults 26.7% 68.9% 69.3% 86.2% 14.0% 55.2% 2) Marred adults 41.7% 84.7% 89.9% 91.9% 8.3% 78.2% 3) Dependent chldren 31.6% 75.9% 84.3% 83.0% 8.9% 62.0% 4) Total 100% 77.7% 83.3% 88.0% 10.0% 66.9% Change 1) Unmarred adults -0.1% -1.4% -1.5% -1.7% * 0.3% -2.3% ** 2) Marred adults 0.5% -0.4% -0.1% -1.0% * -0.4% -1.7% ** 3) Dependent chldren -0.4% ** -0.8% -0.3% -3.5% ** -0.1% -3.1% ** 4) Total -0.7% -0.4% -1.9% ** -0.1% -2.2% ** Source: Urban Insttute analyss of the 1999 and 2002 Natonal Survey of Amerca's Famles (NSAF). a. Alternatve sources of ESI refers to those ndvduals who report havng ESI coverage, but who do not have a current offer of coverage, ether through ther own or an observable famly member's place of employment. Ths ncludes those wth coverage through a former employer or a famly member not present n the household. * p -value between.05 and.1 ** p -value less than or equal to.05

Table 2. Share of Change n Employer-Sponsored Insurance Attrbutable to Each Component (percent) DC W O T A S Share who Share wth Share who Share wth take up alt. ESI work offer offer source a ESI coverage 1) Unmarred adults -2.3% 38.4% 38.8% 35.8% -13.0% 2) Marred adults -1.7% 29.3% -18.9% 68.5% 21.0% 3) Dependent chldren -3.1% 17.8% 5.7% 74.0% 2.4% Share of populaton 4) Total -2.2% 27.8% 7.2% 64.4% 4.2% -3.6% Source: Urban Insttute analyss of the 1999 and 2002 Natonal Survey of Amerca's Famles (NSAF). a. Alternatve sources of ESI refers to those ndvduals who report havng ESI coverage, but who do not have a current offer of coverage, ether through ther own or an observable famly member's place of employment. Ths ncludes those wth coverage through a former employer or a famly member not present n the household.

Table 3. Probablty of Nonwork, Self-Employment, and Other Types of Work, 1999-2002 (percent) Nonwork Selfemployed Other a Total Survey Year: 1999 1) Unmarred adults 21.9% 7.2% 0.6% 29.7% 2) Marred adults 6.3% 8.5% 0.2% 14.9% 4) Dependent chldren 15.4% 7.9% 0.0% 23.3% Survey Year: 2002 1) Unmarred adults 23.8% 7.0% 0.3% 31.1% 2) Marred adults 7.3% 7.8% 0.2% 15.3% 4) Dependent chldren 16.4% 7.5% 0.2% 24.1% Change 1) Unmarred adults 1.9% ** -0.2% -0.3% ** 1.4% 2) Marred adults 1.0% ** -0.7% 0.0% 0.4% 4) Dependent chldren 1.0% -0.4% 0.2% * 0.8% Source: Urban Insttute analyss of the 1999 and 2002 Natonal Survey of Amerca's Famles (NSAF). a. "Other" ncludes those workng temporarly or wthout pay. Such workers were not asked whether ther employers offered health nsurance. * p -value between.05 and.1 ** p -value less than or equal to.05

Table 4. Changes n the Components of Employer-Sponsored Insurance Coverage Rates, < 200% of the Federal Poverty Level, 1999-2002 (percent) Survey Year: 1999 S W O T A C Share who Share wth Share who Share wth take up alt. ESI work offer offer source a Share of populaton ESI coverage 1) Unmarred adults 31.6% 49.0% 50.0% 75.3% 13.2% 31.6% 2) Marred adults 27.8% 67.2% 70.8% 77.8% 15.3% 52.3% 3) Dependent chldren 40.5% 64.9% 68.0% 67.5% 8.2% 38.0% 4) Total 100.0% 60.5% 64.2% 72.6% 11.7% 39.9% Survey Year: 2002 1) Unmarred adults 33.1% 45.2% 50.0% 70.6% 14.1% 30.1% 2) Marred adults 27.7% 64.7% 66.0% 75.3% 13.6% 45.7% 3) Dependent chldren 39.2% 62.6% 65.2% 58.6% 8.5% 32.5% 4) Total 100% 57.5% 61.5% 66.8% 11.8% 35.3% Change 1) Unmarred adults 1.4% ** -3.7% ** 0.0% -4.7% ** 0.9% -1.5% 2) Marred adults -0.1% -2.5% -4.8% ** -2.5% -1.7% -6.6% ** 3) Dependent chldren -1.3% ** -2.3% ** -2.8% ** -8.9% ** 0.4% -5.5% ** 4) Total -3.1% ** -2.7% ** -5.8% ** 0.0% -4.6% ** Source: Urban Insttute analyss of the 1999 and 2002 Natonal Survey of Amerca's Famles (NSAF). Note: Poverty thresholds reflect ncome n calendar years 1999 and 2002. See text for detals. a. Alternatve sources of ESI refers to those ndvduals who report havng ESI coverage, but who do not have a current offer of coverage, ether through ther own or an observable famly member's place of employment. Ths ncludes those wth coverage through a former employer or a famly member not present n the household. * p -value between.05 and.1 ** p -value less than or equal to.05

Table 5. Share of Change n Employer-Sponsored Insurance Attrbutable to Each Component, < 200% of the Federal Poverty Level (percent) DC W O T A S Share who Share wth Share who Share wth take up alt. ESI work offer offer source a Share of populaton 1) Unmarred adults -1.5% 90.6% -0.1% 74.2% -64.7% 2) Marred adults -6.6% 19.5% 36.6% 17.1% 26.9% 3) Dependent chldren -5.5% 17.2% 19.6% 70.0% -6.9% ESI coverage 4) Total -4.6% 26.1% 24.2% 48.1% 0.6% 1.1% Source: Urban Insttute analyss of the 1999 and 2002 Natonal Survey of Amerca's Famles (NSAF). Note: Poverty thresholds reflect ncome n calendar years 1999 and 2002. See text for detals. a. Alternatve sources of ESI refers to those ndvduals who report havng ESI coverage, but who do not have a current offer of coverage, ether through ther own or an observable famly member's place of employment. Ths ncludes those wth coverage through a former employer or a famly member not present n the household.

Table 6. Probablty of Nonwork, Self-Employment, and Other Types of Work, < 200% of the Federal Poverty Level, 1999-2002 Nonwork Selfemployed Other a Total Survey Year: 1999 1) Unmarred adults 43.3% 6.9% 0.9% 51.1% 2) Marred adults 20.1% 12.2% 0.5% 32.8% 4) Dependent chldren 25.8% 9.0% 0.3% 35.1% Survey Year: 2002 1) Unmarred adults 47.2% 7.1% 0.5% 54.8% 2) Marred adults 23.8% 10.6% 0.8% 35.3% 4) Dependent chldren 28.6% 8.3% 0.5% 37.4% Change 1) Unmarred adults 3.9% ** 0.1% -0.4% 3.7% ** 2) Marred adults 3.7% ** -1.6% 0.4% 2.5% 4) Dependent chldren 2.8% ** -0.7% 0.2% 2.3% ** Source: Urban Insttute analyss of the 1999 and 2002 Natonal Survey of Amerca's Famles (NSA Note: Poverty thresholds reflect ncome n calendar years 1999 and 2002. See text for detals. a. "Other" ncludes those workng temporarly or wthout pay. Such workers were not asked whether ther employers offered health nsurance. * p -value between.05 and.1 ** p -value less than or equal to.05

Table 7. Changes n the Components of Employer-Sponsored Insurance Coverage Rates, = 200% of the Federal Poverty Level, 1999-2002 (percent) Survey Year: 1999 S W O T A C Share who Share wth Share who Share wth take up alt. ESI work offer offer source a Share of populaton ESI coverage 1) Unmarred adults 24.4% 83.9% 78.5% 90.8% 14.1% 73.9% 2) Marred adults 47.8% 90.1% 94.1% 95.3% 6.9% 87.7% 3) Dependent chldren 27.8% 85.0% 93.6% 94.1% 9.6% 84.5% 4) Total 100.0% 87.2% 90.3% 94.0% 9.4% 83.5% Survey Year: 2002 1) Unmarred adults 23.7% 84.7% 76.3% 89.8% 14.0% 72.0% 2) Marred adults 48.4% 90.2% 94.6% 94.2% 6.8% 87.1% 3) Dependent chldren 27.9% 84.9% 93.8% 91.4% 9.2% 82.0% 4) Total 100% 87.4% 90.2% 92.6% 9.2% 82.1% Change 1) Unmarred adults -0.7% 0.8% -2.3% * -1.0% -0.1% -1.9% 2) Marred adults 0.6% 0.1% 0.5% -1.1% * -0.1% -0.6% 3) Dependent chldren 0.1% -0.2% 0.2% -2.7% ** -0.4% -2.5% ** 4) Total 0.2% -0.1% -1.5% ** -0.2% -1.3% Source: Urban Insttute analyss of the 1999 and 2002 Natonal Survey of Amerca's Famles (NSAF). Note: Poverty thresholds reflect ncome n calendar years 1999 and 2002. See text for detals. a. Alternatve sources of ESI refers to those ndvduals who report havng ESI coverage, but who do not have a current offer of coverage, ether through ther own or an observable famly member's place of employment. Ths ncludes those wth coverage through a former employer or a famly member not present n the household. * p -value between.05 and.1 ** p -value less than or equal to.05

Table 8. Share of Change n Employer-Shared Insurance Attrbutable to Each Component, = 200% of the Federal Poverty Level (percent) DC W O T A S Share who Share wth Share who Share wth take up alt. ESI work offer offer source a ESI coverage 1) Unmarred adults -1.9% -29.8% 90.3% 33.3% 6.2% 2) Marred adults -0.6% -9.3% -70.3% 166.3% 13.3% 3) Dependent chldren -2.5% 6.0% -7.1% 85.3% 15.8% Share of populaton 4) Total -1.3% -8.9% 12.2% 91.0% 13.2% -7.6% Source: Urban Insttute analyss of the 1999 and 2002 Natonal Survey of Amerca's Famles (NSAF). Note: Poverty thresholds reflect ncome n calendar years 1999 and 2002. See text for detals. a. Alternatve sources of ESI refers to those ndvduals who report havng ESI coverage, but who do not have a current offer of coverage, ether through ther own or an observable famly member's place of employment. Ths ncludes those wth coverage through a former employer or a famly member not present n the household.

Table 9. Changes n the Components of Employer-Sponsored Insurance Coverage Rates, by Establshment Sze, 1999-2002 Survey Year: 1999 S W O T A C Share wth Share who work Share wth offer Share who take up alt. ESI Small Large Gov't Small Large Gov't Small Large Gov't source a Share of populaton ESI coverage 1) Unmarred adults 26.8% 22.8% 37.4% 10.1% 53.9% 77.9% 82.5% 82.2% 88.9% 92.8% 13.7% 57.5% 2) Marred adults 41.2% 16.2% 46.9% 22.0% 71.3% 93.7% 96.0% 93.2% 94.1% 87.7% 8.8% 79.9% 3) Dependent chldren 32.0% 17.0% 42.3% 17.4% 63.2% 89.5% 93.4% 76.1% 88.7% 88.5% 9.0% 65.1% 4) Total 100.0% 18.2% 42.9% 17.3% 63.0% 88.7% 93.1% 82.6% 91.3% 91.6% 69.2% Survey Year: 2002 1) Unmarred adults 26.7% 23.7% 34.0% 11.1% 49.0% 80.1% 79.6% 77.2% 87.9% 92.6% 14.0% 55.2% 2) Marred adults 41.7% 16.6% 45.9% 22.2% 68.7% 94.6% 95.9% 92.2% 93.9% 83.5% 8.3% 78.2% 3) Dependent chldren 31.6% 17.5% 40.5% 17.9% 61.9% 89.9% 93.4% 68.5% 85.7% 86.5% 8.9% 62.0% 4) Total 100.0% 18.8% 41.0% 17.9% 60.1% 89.9% 92.4% 77.2% 90.2% 90.4% 66.9% Change 1) Unmarred adults -0.1% 0.9% ** -3.4% ** 1.0% * -4.9% ** 2.2% -2.9% -5.0% ** -1.0% -0.2% 0.3% -2.3% ** 2) Marred adults 0.5% 0.4% -1.0% 0.2% -2.6% 0.9% -0.1% -1.0% -0.2% -4.2% ** -0.5% -1.7% ** 3) Dependent chldren -0.4% ** 0.5% -1.9% ** 0.6% -1.3% 0.4% 0.0% -7.5% ** -3.0% ** -2.1% -0.1% -3.1% ** 4) Total 0.0% 0.6% -1.9% ** 0.6% -2.9% * 1.2% * -0.7% -5.4% ** -1.1% * -1.2% -2.2% Source: Urban Insttute analyss of the 1999 and 2002 Natonal Survey of Amerca's Famles (NSAF). Notes: Small frms are those wth fewer than 25 workers. For two-earner famles, f at least one earner s employed n government, all famly members are categorzed as government. Otherwse, f at least one worker s employed n a large establshment, all famly members are categorzed as large. All other ndvduals n workng famles are categorzed as small. a. Alternatve sources of ESI refers to those ndvduals who report havng ESI coverage, but who do not have a current offer of coverage, ether through ther own or an observable famly member's place of employment. Ths ncludes those wth coverage through a former employer or a famly member not present n the household. * p -value between.05 and.1 ** p -value less than or equal to.05