More Information Isn t Always Better: The Case of Voluntary Provision of Environmental Quality. Ann L. Owen. Julio Videras.

Size: px
Start display at page:

Download "More Information Isn t Always Better: The Case of Voluntary Provision of Environmental Quality. Ann L. Owen. Julio Videras."

Transcription

1 More Informaton Isn t Always Better: The Case of Voluntary Provson of Envronmental Qualty Ann L. Owen Julo Vderas Stephen Wu Hamlton College September 2008 Abstract Ths paper adds to the lterature on the voluntary provson of publc goods by showng that the warm glow that ndvduals gan depends on the perceved relatve effectveness of contrbutons. We use a new survey on pro-envronment behavors, atttudes, and knowledge and fnd that ndvduals act n accordance wth ther belefs, regardless of whether or not these belefs are accurate, and engage more frequently n actvtes that have a hgher perceved mpact on envronmental qualty. We fnd that low provson of the publc good s greater among people who beleve they cannot do much for the envronment and do not consder themselves envronmentalsts. Ann Owen Julo Vderas Stephen Wu Department of Economcs Department of Economcs Department of Economcs Hamlton College Hamlton College Hamlton College 198 College Hll Road 198 College Hll Road 198 College Hll Road Clnton, NY Clnton, NY Clnton, NY aowen@hamlton.edu jvderas@hamlton.edu swu@hamlton.edu Ths work s supported by a grant from the Blue Moon Fund. Owen also acknowledges support from the Populaton Studes and Tranng Center at Brown Unversty. We thank Emly Conover, Davd Rvera, Joel Shapro, Jennfer Thacher, and semnar partcpants at Brown Unversty for helpful comments.

2 More Informaton Isn t Always Better: The Case of Voluntary Provson of Envronmental Qualty September 2008 Abstract Ths paper adds to the lterature on the voluntary provson of publc goods by showng that the warm glow that ndvduals gan depends on the perceved relatve effectveness of contrbutons. We use a new survey on pro-envronment behavors, atttudes, and knowledge and fnd that ndvduals act n accordance wth ther belefs, regardless of whether or not these belefs are accurate, and engage more frequently n actvtes that have a hgher perceved mpact on envronmental qualty. We fnd that low provson of the publc good s greater among people who beleve they cannot do much for the envronment and do not consder themselves envronmentalsts.

3 1 Introducton The lterature on the voluntary provson of publc goods shows that contrbutons are larger than would be expected f ndvduals were purely self-nterested. In Andreon s theory of mpure altrusm, contrbutng creates a warm glow. Accordng to ths theory, t s the act of gvng that generates utlty (Andreon, 1989, 1990). Duncan (2004) proposes an alternatve motvaton, mpact phlanthropy, where the ncrease n the publc good caused by the ndvdual s efforts creates utlty, whle Brekke, Kverndokk and Nyborg (2003) develop a model n whch ndvduals gan utlty when they act n accordance wth ther self-mage as socallyresponsble people. Our paper extends ths lterature by developng a model n whch the perceved effectveness of effort generates utlty but ndvduals mght have naccurate and dfferng nformaton about the mpact of specfc efforts. We then present evdence from a new natonally representatve household survey that supports the key assumptons and conclusons of the model. In partcular, n the case of the publc good of envronmental qualty, we show that most ndvduals hold ncorrect belefs about the mpact of ther pro-envronment efforts, but efforts correlate consstently wth perceved mpacts. These results suggest that t s necessary to consder the lack of accurate nformaton n explanng ndvdual contrbutons to publc goods. Examnng the relatonshp between the perceved mpact of contrbutons and actual contrbutons has mportant mplcatons for publc polcy. Frst, although t s temptng to conclude that more educaton about the value of ndvdual efforts to ncrease the publc good would result n greater provson, our model and emprcal results mply that the effect of more educaton s ambguous. As ndvduals develop a better understandng of the effectveness of varous actvtes, the quantty of the publc good that they voluntarly provde could ether ncrease or decrease, dependng n part on whether they ntally underestmate or overestmate

4 the relatve mpact of ther efforts. In what follows, we dscuss the crcumstances under whch educaton may lead to a more effcent allocaton of contrbutons and the condtons under whch educaton may actually lead to less of the publc good. A second polcy concluson, ndependent of whether ndvduals are fully nformed or not, s that government regulaton would not be a perfect substtute for voluntary pro-envronment actons. Ths result parallels the concluson n Andreon (1989). Whle our model can be appled to dfferent publc goods, we examne emprcally one of today s foremost publc polcy challenges: reducton of greenhouse gases or, more generally, resource conservaton. A great amount of nformaton and advocacy efforts are beng dedcated to these ssues but t s unclear whether these campagns are successful and whether ndvduals are wllng to make sustanable consumpton choces voluntarly. 1 We use data from a new representatve natonal survey and estmate the relatonshp between the perceved mpacts of several actvtes on emssons of carbon doxde and actual frequency of pro-envronment behavors. We fnd that, on average, ndvduals overestmate the effectveness of ther efforts on emssons of carbon doxde but people act accordng to ther perceptons and engage more frequently n actvtes that have hgher perceved mpacts. Thus, we present evdence for a modfed warm glow, a warm glow based on the perceved mpact of contrbutons rather than the contrbuton tself. Ths paper relates to several research areas. Frst, t adds to the lterature analyzng the role that nformaton plays n publc goods contrbutons. Kremer and Mguel (2007) nvestgate the determnants of ndvdual actons to prevent the spread of nfectous dsease n Kenya and fnd that school health educaton programs had no effect on ndvdual behavor. Usng 1 A popular example s the publc campagn assocated wth the flm An Inconvenent Truth. In fact, we use nformaton on ths flm s web ste, as the bass for some of the examples we provde n our survey. 2

5 expermental data, Andreon (1995) examnes whether publc goods contrbutons can be attrbuted to kndness or confuson and fnds evdence for both. Houser and Kurzban (2002) present corroboratng evdence. In these papers, the confuson s specfc to the expermental desgn. In our emprcal work, we fnd that ndvduals hold naccurate belefs about the mpact of actvtes n whch they regularly engage, but the level of ther contrbutons s stll consstent wth those belefs. Ths paper also adds to the lterature on the motves for altrusm. As mentoned earler, Duncan (2004) develops a model n whch ndvduals gan utlty from the ncrease n the publc good caused by ther efforts. Our model takes nto account Duncan s mpact phlanthropy yet also ncludes Andreon s orgnal warm glow motve because ndvduals gan utlty from the perceved mpact of the effort, not from the actual consumpton of the publc good. 2 More mportantly, we allow for ndvduals to have naccurate nformaton about the effectveness of ther contrbutons, an extenson that has both emprcal and polcy relevance. Also related to our work s the research by Brekke, Kverndokk, and Nyborg (2003) examnng whether a desre to be socally responsble motvates recyclng efforts and communty work. As we do, they consder the effectveness of efforts n ther model. However, they focus on how effort relates to self-mage and do not consder the possblty that ndvduals may have dfferent levels of accuracy of nformaton. 3 An mportant mplcaton of Brekke, Kverndokk, and Nyborg (2003) s that publc polcy mght decrease the prvate provson of the publc good f mandated behavor makes t more dffcult for ndvduals to fulfll ther deal actons. 4 Our 2 In other words, n our model, ndvduals gan utlty from the efforts even when they beleve that ther acton has no apprecable effect on the total quantty of the publc good. 3 In ther model, allowng ndvduals to have dfferent perceptons about the effectveness of effort would pose addtonal complcatons unnecessary for ther man pont. In partcular, t would then be necessary to propose a mechansm through whch ndvduals would defne and learn the amount of effort a socally responsble person ought to exert. 4 See also Bruvoll and Nyborg (2004) and Nyborg and Rege (2003). 3

6 research also suggests publc polcy mght have unexpected consequences, although n our case ths concluson s due n part to the possblty that nformaton about the mpacts of dfferent pro-envronment behavors could make ndvduals revse downwards the perceved effectveness of ther actons and provde less of the publc good. Fnally, our work relates to the lterature examnng values and sustanable consumpton. 5 Ths area of research usually focuses on the categorzaton of values (for example, unversalsm versus ndvdualsm) and ther nfluence on atttudes, ntentons, and behavors. Although we do control for values n our analyss, our man hypotheses relate to the effects of perceved effectveness of efforts on the type and ntensty of such efforts. The paper proceeds as follows. We present the theoretcal framework n Secton 2. Secton 3 presents the emprcal methods and man hypotheses. Secton 4 descrbes the survey desgn and the orgnal data set used for our emprcal analyss. Secton 5 dscusses the man results and robustness checks and secton 6 concludes. 2 Theoretcal Framework 2.1 Utlty Maxmzaton We model ndvduals who can contrbute to a publc good through J dstnct actvtes. The utlty of ndvdual s equal to: U u x, G, g ( e ),... g ( e )) (1) ( 1 1 J J where x represents consumpton of the prvate good, G represents the total amount of the publc good, e j represents the effort that ndvdual makes n performng actvty j, (j=1 to J), and g j (e j ) represents ndvdual s contrbuton to the publc good through actvty j. We assume utlty s ncreasng and concave n all arguments, so that u ( ) 0, u( ) 0. Tme s spent ether 5 The lterature examnng values and sustanable consumpton s large. See, among many others, Thogersen and Olander (2002), and Detz, Gergory, and Guagnana (1998). 4

7 producng the prvate good or supplyng efforts e j to a partcular actvty. We assume that an ndvdual s contrbutons to the publc good through actvty j ncreases wth efforts at a decreasng rate and allow for g j (e j ) to vary across ndvduals wth g ( 0) 0, g '( ) 0, g ''( ) 0 for all j. 6 Tme spent n the prvate sector s drectly converted j j j nto the prvate good so that ndvduals face the followng tme constrant: x e T (2) j j The total amount of the publc good s a lnear functon of the mpact of the ndvdual efforts: G G( g( e)) b g j ( e j ) (3) j where b>0 s a constant. Substtutng equatons (2) and (3) nto equaton (1) gves: U ut ej, bg j( ej), g1( e1),.. gj ( ej ) (4) j j Frst, we solve for the socal optmum, takng nto account the warm glow motve of ndvduals. The socal optmum s acheved by maxmzng The frst order condtons are: u T ej, bg j( ej), g1 ( e1),.. gj ( ej ) (5) j j U x g e j j b U g U j, j G ej g j (6) 6 Ths assumpton of decreasng returns s reasonable f ndvduals frst undertake actvtes that are the easest n contrbutng to the publc good. Abatement mght ncrease wth effort at an ncreasng rate f there were learnngby-dong. 5

8 For each ndvdual, there are J frst order condtons that correspond to the J possble actvtes that contrbute to the publc good. Each FOC states that the margnal utlty of the prvate good s set equal to the sum of two terms. The frst term s the margnal product of effort expended through actvty j tmes the margnal product of the ndvdual s contrbuton n ncreasng the publc good tmes the sum of the ndvdual margnal utltes of the publc good. The second term s the margnal product of effort tmes the margnal utlty of the ndvdual s contrbuton to the publc good. Ths second term represents the modfed warm glow effect, a warm glow that depends on the mpact of effort and not exclusvely on the amount of effort. 2.2 Voluntary Provson Now we derve the condtons for voluntary provson of the publc good. Indvdual chooses effort levels e 1, e J to maxmze equaton (4). The frst order condtons are: U x g e j j U b G U g j, j (7) Notce the dfference between the FOC s n equaton (6) and the FOC s n equaton (7). To acheve the socal optmum, the frst term on the rght hand sde n equaton (6) ncludes the sum of each ndvdual s margnal utlty of the publc good. Prvate provson of the publc good wll yeld FOC s where the frst term on the rght hand sde ncludes only that ndvdual s margnal utlty of the publc good. Thus, we obtan the standard result that too lttle of the publc good s produced relatve to the socal optmum. 2.3 Perceptons and Behavor In our emprcal models we examne how perceptons about the effectveness of specfc actvtes that reduce emssons of carbon doxde correlate wth the frequency wth whch ndvduals undertake pro-envronment behavors. Thus, we can nterpret g () as the technology j 6

9 that transforms effort nto carbon abatement and G () as the technology for transformng carbon reductons nto envronmental qualty. Although there s an actual technology g () for each actvty j, ndvduals mght not accurately assess the effectveness of ther efforts. As we descrbe n the next secton, we fnd that n a natonally representatve sample, ndvduals dffer n ther belefs about how several actvtes reduce emssons of carbon doxde and that typcal respondents msperceve the mpact of ther efforts. It s also the case that ndvduals hold dfferent belefs about technology that creates the publc good (n our model ths s Equaton 3). For example, n our sample, only 22 percent of the respondents say t s defntely true that usng coal, ol, or gasolne contrbutes to clmate change. j Let gˆ j (.) be the belef of ndvdual about the effectveness of actvty j n reducng carbon emssons and let bˆ be the belef of ndvdual about the technology that transforms carbon reducton nto envronmental qualty. Msperceptons can occur when ˆ b and g ˆ ( ) g j j ( ). An ndvdual wthout perfect nformaton chooses the optmal level of efforts gven gˆ ( ) rather than the actual technology. Thus, the frst order condton s: j b U x gˆ e j j U bˆ Gˆ U gˆ j, j (8) An examnaton of Equaton 8 reveals that wthout makng further restrctve assumptons, the effect of more accurate nformaton s ambguous. The reason for ths s that changes n the perceved effectveness of efforts have both an output effect and a substtuton effect. For example, as ndvduals who ntally underestmate the mpact of ther efforts become better nformed, the output effect decreases the optmal level of effort because 7

10 ndvduals are able to contrbute more to the publc good wth less effort. On the other hand, the substtuton effect ncreases effort because ndvduals allocate more tme to actvtes wth relatvely hgher mpacts. If the output effect domnates, then when ndvduals learn that actvty j has a hgher mpact than orgnally perceved, they wll decrease the tme they spend n that actvty. If the substtuton effect domnates, however, ndvduals wll put more effort nto that actvty. The analogous conclusons hold when ndvduals overestmate the mpact of ther efforts they may ncrease or decrease ther efforts when they learn the truth. Because we do not have theoretcal grounds on whch to mpose assumptons to determne unambguously the optmal response on efforts, how ndvduals respond to perceved mpact of ther actons becomes an emprcal queston whch we take up n the remander of the paper. The next secton descrbes our emprcal approach for provdng evdence on ths ssue. In partcular, we emprcally estmate the effect of belefs about mpact of efforts on the amount and type of effort that ndvduals exert. Secton 4 presents the data and provdes more specfcs on the varables used n the analyss. Secton 5 provdes results that ndcate ndvduals are n fact more lkely to engage n pro-envronment behavors the hgher the perceved effects of specfc actvtes on carbon emssons. 3 Methods and Hypotheses To gather evdence on the role that perceved mpacts have on pro-envronment behavors, we examne the factors affectng four dfferent pro-envronment behavors. Specfcally, our dependent varables measure the frequency over the past 12 months wth whch ndvduals undertake each of four behavors out of concern for the envronment: recyclng, reducng energy consumpton at home, buyng envronmentally frendly products, and alterng food consumpton. We do not observe the actual amount of effort ndvduals dedcated to each 8

11 actvty. Rather, we observe responses on a 1 to 4 scale wth 1 correspondng to never, 2 correspondng to occasonally, 3 correspondng to frequently, and 4 correspondng to nearly all the tme. Thus, our dependent varables are ordnal. Because these varables volate the assumpton of the lnear regresson model of equal dstance between categores, we present results from ordered probt models. 7 For a gven actvty and ndvdual, the actual level of effort y* s unobserved. In our emprcal model, y* depends on soco-economc characterstcs, general atttudes, values, and knowledge, and perceved effectveness of specfc behavors on carbon abatement. Lettng the vector X represent these controls, the structural model for ndvdual s gven by: y. We observe y =1 f y * 1, y = 2 f 1 y * 2, y = 3 f 2 y * 3, * X and y = 4 f 3 y*, where y can be our ordnal measure of the frequency of recyclng, energy conservaton, use of envronmentally frendly products, or alterng food consumpton. The parameters are thresholds such that the observed response changes as the unobserved level of effort y* crosses the cut-off ponts. For gven values of the ndependent varables, the probablty of outcome m (m = 1 to 4) s: Pr( y m X ) F( X ) F( 1 X ), where we assume F s the normal cumulatve densty functon wth Var(ε) = 1. 8 Equaton 8 provdes the frst order condtons to the ndvdual s problem and we use that equaton to gude our emprcal work. As we descrbe n more detal below, we are able to proxy m m for ˆ ˆ ˆ gj ej, G gj, U g j and ˆ U Gˆ wth responses to questons from a natonally representatve household survey. In the next secton, we descrbe ths survey n more detal and dscuss how the data s used to proxy these concepts. 7 We also estmate ordered logt models, multnomal logt and probt models, and OLS models. We fnd the results are robust to the estmaton method. We dscuss these ssues n more detal n Secton 5. 8 We replcate all models assumng F s logstc. We draw the same man nferences but the standard errors under the assumpton of normal errors are systematcally smaller n our data. 9

12 4 Survey Desgn and Data To test the hypotheses of the model we use data for approxmately 1,700 respondents from a new natonally representatve household survey conducted n September and October of The respondents to the survey were part of the Knowledge Networks Internet panel who were recruted va random dgt dalng. Knowledge Networks uses a unque sample desgn for Internet panels, provdng households Internet access to avod the based sample that results from requrng partcpants to obtan Internet access on ther own. 9,10 Volunteer panelsts are not accepted by Knowledge Networks. The survey nstrument contaned ffty questons. 11 The frst set of questons asked about general atttudes toward the natural envronment. The second group elcted how frequently ndvduals engage n pro-envronment behavors out of concern for the envronment. Thrd, the survey evaluated the respondents general knowledge of envronmental problems and belefs about the effectveness of specfc actvtes on emssons of carbon doxde. Fnally, the survey asked questons about tme preferences, rsk averson, and atttudes towards free rdng. We augment the survey wth respondent demographcs and an array of ndvdual characterstcs that Knowledge Networks collects as part of ther publc affars profle, a seres of questons that are asked perodcally of all members of the panel. To measure contrbutons to the publc good of envronmental protecton we use four questons that elct how frequently ndvduals engage n pro-envronment behavors out of 9 Internet surveys have several advantages. They allow for more complex questons than can be asked n a telephone survey and are less lkely to be subject to ntervewer bas (tryng to please the ntervewer by respondng the rght way ) than telephone or face-to-face surveys are. See Krosnck and Chang (2001) for a comparson of random dgt dalng telephone ntervews, the Knowledge Networks Internet panel, and other Internet panels. 10 The response rate among Knowledge Networks panelsts for our survey was 66%. Berrens et al. (2004) also present results usng a survey mplemented by Knowledge Networks (KN) on wllngness to pay for clmate change mtgaton and Cameron and DeShazo (2001, 2004) show that ther KN sample s comparable to data from the 2000 Census. 11 The entre survey as well as more detaled nformaton about the survey methodology can be obtaned from 10

13 concern for the envronment. We focus on recyclng (RECYCLE), reducng energy consumpton at home (ENERGY), buyng envronmentally frendly products (PRODUCT), and alterng food consumpton (FOOD). The survey gave some specfc examples for the behavors such as: washng clothes n cold water nstead of hot as a way to reduce energy consumpton, usng energy-savng lght bulbs as an example of buyng envronmentally frendly products, and eatng less meat as a way to alter food consumpton for envronmental reasons. A man result of the model s that efforts to reduce one s carbon footprnt through varous actvtes should correlate wth the perceved effectveness of the actvtes. A unque aspect of our survey s a seres of questons that assess the respondents belefs about the effectveness of specfc behavors on mprovng envronmental qualty. Frst, the survey noted that scentsts thnk that average global temperatures are rsng and global clmate s changng because carbon doxde from burnng coal and ol and other greenhouse gases are released nto the atmosphere. 12 Then, as a baselne comparson, the survey stated that adjustng the thermostat n a typcal household up two degrees n the summer and down two degrees n the wnter s assocated wth a 2,000 pound reducton of carbon emssons per year. Four subsequent questons asked the respondent to rate the mpact of dfferent actvtes n terms of the amount of carbon reducton: recyclng half of household garbage (RECYCLE_BELIEF), usng cold water nstead of hot to wash one s clothes (COLD_BELIEF), replacng fve regular lght bulbs wth compact fluorescent lght bulbs (LIGHT_BELIEF), and elmnatng all anmal products from one s det (VEGAN_BELIEF). Respondents were asked to make ther best guess as to how much these behavors reduced carbon emssons per year: sgnfcantly less than adjustng your thermostat (less than 1,500 pounds), about the same as adjustng your thermostat (1,500 up to 12 Ths nformaton was provded after respondents answered questons elctng general knowledge about causes of clmate change. 11

14 2,500 pounds), and sgnfcantly more than adjustng your thermostat (more than 2,500 pounds). We code these responses n two ndcator varables for each behavor (where the omtted category s less than 1,500 pounds). We also create an ndex from these four questons by gvng each respondent one pont for each correct answer, SCORE. In the estmaton of pro-envronment behavors, we also control for general knowledge about envronmental problems as well as atttudes and values related to the envronment, as these correlate wth behavors and perceptons of the effectveness of specfc actvtes. The survey asked ndvduals whether or not they consdered themselves to be an envronmentalst. From the responses to ths queston, we constructed two ndcator varables, GREEN_SOME and GREEN_DEF, ndcatng those who responded yes, somewhat and yes, defntely, respectvely. 13 Our respondents were asked the same queston approxmately sx months before they completed our survey, as part of Knowledge Networks publc affars profle. We repeated the queston to explore whether ndvduals mght want to appear to have the rght atttudes n our survey. We fnd a hgh degree of correlaton between the responses: only seven people who sad they were defntely not an envronmentalst sx months earler clam to defntely be an envronmentalst n our survey. Ths consstency across tme and n dfferent contexts adds confdence to our data. We nclude three varables that measure basc knowledge about clmate change by usng the extent to whch people beleve t s true that Every tme we use coal, ol, or gas, we contrbute to clmate change. Those who sad ths statement was defntely true are ndcated by 13 We also expermented n our models wth commonly used envronmental atttudes and found those varables are statstcally nsgnfcant after controllng for self-reported envronmentalsm. Those results suggest that our measure of envronmentalsm summarzes well a person s overall atttudes toward the envronment. 12

15 the ndcator varable, COAL_DEF, those who sad t was probably true are ndcated by COAL_PROB, and those who sad t was probably not true are ndcated by COAL_NOT. 14 To control for the degree to whch ndvduals beleve ther actons nfluence overall envronmental qualty, we nclude the varable FATALIST that equals one f the ndvdual strongly agrees or agrees that t s dffcult for somebody lke me to do much about the envronment. Indvduals mght also contrbute to the publc good f the level of the publc good tself generates utlty. On a scale of one to four, PERSONAL ndcates the extent to whch people beleve that clmate change wll affect them personally and LIVSTAND ndcates the extent to whch people beleve that envronmental damage wll cause a reducton n lvng standards n the next 50 years. We nclude two varables related to an ndvdual s overall propensty to contrbute to publc goods: a proxy for socal responsblty and a proxy for optmsm. We measure socal responsblty by summng the responses to questons about the justfablty of cheatng on taxes, rdng publc transportaton wthout payng the fare, downloadng copyrghted musc or moves wthout permsson, and buyng stolen goods. Respondents state on a scale of one to ten where a ten ndcates that the behavor can never be justfed whle a one ndcates that the behavor s always justfable. 15 The sum of these responses becomes an ndex of cvc responsblty, CIVIC, whch ranges from 4 to We also control for an ndvdual s overall level of optmsm by ncludng the response to a queston that elcts, on a scale of one to four, how strongly ndvduals agree wth the statement that the U.S. economy wll mprove n the next fve years. 14 The survey ncluded other general knowledge questons but ths one has the strongest predctve power. 15 These questons and scales of responses are smlar to ones that appear n the World Values Survey. 16 Ths treatment parallels that n Knack and Keefer (1997) who use a smlar set of questons from the World Values Survey to measure cvc responsblty at the country level. Only about one thrd of our sample ndcated that all of these behavors are never justfable. 13

16 Demographc controls nclude dchotomous varables for marred respondents, homeowners, Afrcan-Amercans, Hspancs, and two varables ndcatng f the ndvdual s a hgh school or a college graduate. We also nclude age and age squared, self-reported health status, the log of household ncome (at the census block level), and the fracton of the populaton n the respondent s zp code that s classfed as beng n an urban area. These demographc varables control for the opportunty costs of engagng n pro-envronment actvtes. For example, more educated ndvduals may be more senstve to envronmental ssues or respondents wth hgher ncome may fnd t easer to ncur costs assocated wth pro-envronment actons such as buyng more envronmentally-frendly products. Recyclng could be more convenent for those who lve n urban areas or who own ther own homes. Fnally, we account for geographcal factors that can nfluence the opportunty cost of engagng n the behavors wth ndcator varables for regon (Northeast, Mdwest, South, and West). 17 In Secton 5.3, we dscuss addtonal results when we nclude state-level varables, n partcular, average retal prces of electrcty, proporton of a state s populaton wth access to curbsde programs, and average prces receved by beef cattle farmers. Descrptve statstcs and defntons for these varables appear n Table 1. Compared to the U.S. Census Bureau demographc statstcs, both our unweghted and weghted data are representatve of the U.S. populaton. The 2000 Census estmates that the U.S. populaton s 51 percent female, 12 percent Afrcan-Amercan, and 11 percent Hspanc, all wthn a 95 percent confdence nterval for the means n our data. Nonetheless, we use weghted data for Table 1 and all models We have also estmated the models consderng nne, rather than four, geographcal regons. The results for the varables of nterest are very smlar and we present the results of the more parsmonous model. 18 Demographc and geographc dstrbutons from the Current Populaton Survey as well as nformaton from the entre Knowledge Networks panel re Internet access are used as benchmarks n the constructon of the weghts. See 14

17 Nearly all the tme s the modal response for RECYCLE at approxmately 45 percent of the sample. Frequently s the modal response for ENERGY and PRODUCT (39 and 38 percent, respectvely) whle Occasonally s the most frequent response for FOOD (39 percent of the sample). There are 173 unque response patterns to these questons and the frequency of patterns s very evenly dstrbuted. Overall, there s a substantal amount of varablty n the frequency wth whch ndvduals engage n these four behavors. Regardng how people perceve the effectveness of several actvtes, we fnd that most people do not have an accurate sense of the mpact of ther actons. The source of error s that, on average, respondents overstate the effectveness of some actvtes. Ths can explan why fatalsts do slghtly better on the total score for ths four queston quz. It s nterestng to compare other statstcs between those who beleve that they cannot do much about the envronment (fatalsts) and those who thnk they can. As Table 1 shows, non-fatalsts are more lkely to engage n all types of behavor, to descrbe themselves as envronmentalsts, and rate the ndvdual actvtes (recyclng, usng cold water, usng compact fluorescent lght bulbs, becomng a vegan) as havng a hgh mpact. On the other hand, fatalsts and non-fatalsts seem to be equally cvc-mnded as ndcated by the averages for the ndex of cvc behavor, CIVIC. If ths varable s related to an ndvduals ablty to receve a pure warm glow (a beneft from effort, regardless of the mpact), ths smlarty could explan why the fatalsts stll contrbute. Pror to presentng our results, we relate the varables dscussed above to mportant concepts n the frst order condtons of the optmzaton problem that appear n Equaton 8. Specfcally, we control for the perceved effect of varous actvtes on reducng ndvdual carbon emssons, gˆ j e j, wth the specfc belef questons: RECYCLE_BELIEF, for more detal on the calculaton of the weghts. Our man conclusons are unaffected by the use of samplng weghts. Results for estmatons wthout weghts and any others dscussed but not reported n detal are avalable from the authors upon request. 15

18 COLD_BELIEF, LIGHT_BELIEF, and VEGAN_BELIEF. Recall that our model predcts that the larger the mpact an ndvdual attrbutes to an actvty, the more lkely t s that the ndvdual undertakes that actvty more often. Ths mples postve coeffcent estmates for these varables. After controllng for belef about the mpact of specfc actvtes on carbon abatement, we measure the perceved margnal effect of carbon abatement on the publc good, Gˆ ˆ g j, wth FATALIST. Holdng everythng else constant, we expect the coeffcent on FATALIST to be negatve, lower values of Gˆ gˆ j should be assocated wth less effort n provdng the publc good and greater consumpton of the prvate good. We proxy for the margnal effect of contrbuton on utlty, U gˆ j, wth GREEN_SOME ( somewhat of an envronmentalst ) and GREEN_DEF ( defntely an envronmentalst ). We hypothesze that the coeffcents on these two varables are postve as envronmentalsts should derve more utlty from contrbutng to the publc good of resource conservaton. In addton, CIVIC mght measure overall ncentves to contrbute to publc goods. We control for the margnal effect of the publc good on utlty, U Gˆ, wth PERSONAL and LIVSTAND. We expect the coeffcent on these varables to be postve as utlty should ncrease wth prvate benefts. The model ndcates that an ndvdual s optmal provson of the publc good depends on the nteracton of fatalsm, warm glow, and perceved effectveness. Because we use bnary varables to measure all these effects and the models are farly complex, rather than addng nteracton terms we estmate models for strong and weak envronmentalsts as well as fatalsts and non-fatalsts separately. 16

19 5 Results In ths secton we frst present the results of base specfcatons, dscuss how perceved effectveness of specfc actvtes nfluences efforts, and then check for the robustness of our results. 5.1 Base Models Table 2 presents coeffcents from an ordered probt estmaton when we nclude all varables except the perceved effectveness of specfc actvtes n reducng carbon emssons. Some demographc controls consstently explan the frequency of pro-envronment behavors. Women are more lkely to say that they conserve energy, buy envronmentally frendly products, and alter ther food consumpton out of concern for the envronment than men are. We calculate that women are 6 percent more lkely to say they conserve energy at home almost all the tme than men are. Afrcan-Amercans are 14 percent less lkely to recycle and 10 percent less lkely to conserve energy at home at least frequently than ndvduals of any other race or ethncty, everythng else equal. Those who lve n more urban areas are more lkely to recycle (perhaps because recyclng programs are more wdely avalable to urban resdents) but they are less lkely to report conservng energy or buyng envronmentally frendly products. The more strongly ndvduals agree wth the statement that envronmental degradaton wll cause lvng standards to declne, the more lkely t s that they conserve energy at home (about 5 percent more lkely to do ths actvty nearly all the tme) and alter food consumpton (about 7 percent more lkely to do ths actvty at least frequently). 19 Indvduals who strongly agree wth the statement that clmate change may affect them personally are more lkely to buy envronmentally frendly products and alter food consumpton (the coeffcent n the energy model s sgnfcant at the 10 percent level). The margnal effects are approxmately of the same 19 We obtan qualtatvely smlar results when we enter three ndcator varables for PERSONAL and LIVSTAND. 17

20 magntude as the effects for LIVSTAND. We also fnd the expected sgn for CIVIC, suggestng that those who are more cvc-mnded are more lkely to engage n all of these behavors, ndependent of ther values and belefs about the envronment. 20 As expected, self-proclamed envronmentalsts are more lkely to engage n all behavors. The more defnte ndvduals are about ther envronmentalsm, the larger the effect s. Strong envronmentalsts are almost 29 percent more lkely to conserve energy at home nearly all the tme than non-envronmentalsts, 42 percent more lkely to recycle nearly all the tme than non-envronmentalsts, 36 percent more lkely to purchase envronment-frendly products, and 26 percent more lkely to alter ther food consumpton. The margnal effects for the weak envronmentalsts (relatve to non-envronmentalsts) are approxmately half of the effects for the strong envronmentalsts. Meanwhle, ndvduals who do not beleve that they can have an mpact on the envronment are less lkely to engage n all behavors. Everythng else equal, fatalsts are about 10 percent less lkely to recycle and buy envronment-frendly products nearly all the tme than non-fatalsts. Fatalsts are approxmately 8 percent less lkely to conserve energy and alter food consumpton as often as non-fatalsts. 21 Knowledge that usng coal, ol, or gas contrbutes to clmate change affects recyclng behavor and energy conservaton only. It mght be that t s more dffcult for ndvduals to relate the use of envronmentally frendly products or food consumpton to carbon emssons than t s to understand the relatonshp between recyclng or energy conservaton and carbon emssons. Fnally, these models nclude SCORE, the score that ndvduals receved on the four queston quz about the mpact of specfc behavors on carbon abatement. We fnd that better 20 Ths result corroborates the fndngs of Owen and Vderas (2006) who fnd a smlar effect usng data from the World Values Survey. 21 A number of studes n economcs and other felds show fatalsm to be a strong predctor of behavor such as dsaster preparedness (McClure, Allen and Walkey, 2001), votng behavor (Goodwn and Allen, 2000), and savng (Wu, 2005). 18

21 knowledge about the effect of specfc behavors s negatvely and sgnfcantly related to energy conservaton at the 5 percent level and buyng envronmentally frendly products at the 10 percent. Snce ndvduals tend to overestmate the effectveness of specfc actvtes, ths result s consstent wth a modfed warm glow: a hgher score mples the ndvdual s less lkely to overestmate the mpact of the actvtes and therefore s less lkely to engage n these behavors, all else constant. In the models n Table 2, the estmates of the cut-off ponts are all statstcally dfferent from zero. We also have evdence to reject the null hypothess that the dfference between consecutve thresholds s zero. Thus, the responses ( Almost all the tme, Frequently, Occasonally, and Never ) reflect dstnct meanngful thresholds approxmatng the ntensty of the behavor. Ths suggests t s not approprate to collapse responses nto a bnary ndcator. Although many surveys such as the World Values Survey use dchotomous choce questons, we fnd nterestng results regardng both the type of behavor and the ntensty of the behavor The Effects of Perceved Effectveness of Specfc Actvtes In Table 3 we drop the varable SCORE and add the perceved mpacts of specfc behavors ndvdually. We remnd the reader that we gave the baselne example that adjustng the thermostat up or down two degrees reduces carbon emssons by approxmately 2,000 pounds per year and then asked people to provde ther best guess for the annual reducton n carbon emssons (less than 1,500 pounds, 1,500 to 2,500 pounds, or more than 2,500 pounds) for each actvty: recyclng half of the household s waste (RECYCLE_BELIEF), replacng fve regular lght bulbs wth fve compact fluorescent lght bulbs (LIGHT_BELIEF), elmnatng all other anmal products from det (VEGAN_BELIEF), and usng cold water nstead of warm or hot 22 We also estmate a multnomal logt model and perform a lkelhood-rato test that strongly rejects the null hypothess that any par of categores can be collapsed. 19

22 water to wash clothes (COLD_BELIEF). Thus, we nterpret the coeffcents as the effect of belevng a gven actvty reduces carbon emssons by 1,500 to 2,500 pounds or more than 2,500 pounds relatve to the omtted category (the actvty reduces carbon emssons by less than 1,500 pounds per year). 23 In the frst column of Table 3 we nclude ndcator varables for ndvduals who beleve that recyclng has a medum mpact on reducng carbon emssons and a ndcator varable for those who thnk the mpact s hgh (the correct answer s medum mpact). The postve and sgnfcant coeffcent on the hgh mpact belef (RECYCLE_BELIEF_HI) ndcates that people who beleve that recyclng half of one s household garbage reduces carbon emssons by more than 2,500 pound of carbon emssons per year are more lkely to recycle more often. Columns two through four present smlar fndngs for ENERGY, PRODUCT, and FOOD. Those who beleve that a specfc actvty has a hgher mpact than the baselne are more lkely to engage n the behavor most closely related to that actvty wth greater frequency. Because t mght be possble that people who beleve that all actvtes are hgh mpact are more lkely to engage n each behavor, we nclude ndcator varables for the perceved effectveness of all actvtes smultaneously n each model (columns 5 through 8 of Table 3). We fnd that t s the perceved effect of the actvty most closely assocated wth each behavor that enters sgnfcantly and not the perceved effects of any of the other three actvtes. Table 4 presents the margnal effects from the coeffcent estmates n Table 3, columns 1 through 4. For example, the second column of Table 4 shows the margnal effect of belevng 23 When answerng these questons, ndvduals mght have focused on the relatve comparson wth energy conservaton snce the baselne uses adjustng the thermostat as an example. If ths were the case, we should not fnd that a perceved hgh mpact of usng cold water correlates wth overall energy conservaton. However, we do fnd postve and strongly sgnfcant effects of ths belef on energy conservaton. In addton, we do not fnd that perceved hgh mpacts of other actvtes have a negatve and sgnfcant effect on energy conservaton. These results suggest that ndvduals evaluate the effectveness of the actvtes relatve to the baselne of 2,000 pounds per year and not relatve to the actvty of energy conservaton as a whole. 20

23 that recyclng half of a household s garbage reduces carbon emssons by more than 2,500 pounds per year. Ths percepton reduces the probablty that people never recycle by 2.5 percent, the probablty that people occasonally recycle by 4.7 percent, and the probablty that people frequently recycle by about 1 percent (conversely, t ncreases the probablty that people report recyclng nearly all the tme by 8.1 percent). Smlar fndngs are evdent wth the remanng belefs and behavors. The stronger the perceved effectveness of a specfc actvty s, the hgher the probabltes of engagng n the four pro-envronment behavors wth greater frequences. To strengthen our confdence n these results we perform addtonal analyses on the responses to the questons about perceved mpacts. Frst, t s possble that there s collnearty between the perceved mpacts of varous actvtes. In that case, t mght be hard to determne whether belefs about specfc actvtes are correlated wth behavors. To examne ths ssue we estmate models that nclude each ndvdual knowledge queston separately for each of the behavors. We fnd very lttle sgnfcance for the cross-effects. For example, the only specfc knowledge queston that s statstcally sgnfcant n the recyclng behavor regresson s the one related to the mpact of recyclng. 24 Overall, these results provde evdence that t s the ndvdual s belef about the mpact of specfc actvtes, correct or ncorrect, that drves that partcular behavor. Second, we note that the questons about the perceved effectveness of dfferent actvtes can be dffcult to answer. In that case, people may choose a neutral answer. We examned the response patterns to those four questons and found that the most common pattern, 24 There are a few exceptons: people who beleve recyclng has a large mpact on reducng carbon emssons are also more lkely to report buyng envronmentally frendly products and alterng food consumpton, and people who beleve that usng cold water nstead of hot has a large mpact are also more lkely to report alterng food consumpton. 21

24 about 10 percent of the sample, s to say each actvty has medum mpact. Ths s a potental focal pont. To asses f ths affects our results, we created an ndcator varable that equals 1 f the respondent chooses the most common pattern. Ths varable does not sgnfcantly predct any of the four dependent varables and the estmates of the perceved effect questons and other controls are almost dentcal. Thrd, t mght be possble that the perceved effectveness of dfferent actvtes does not nfluence people s efforts and that, when asked to guess how effectve a gven actvty s, ndvduals assgn greater effectveness to the actvtes they engage n more often. In that event, the responses to the mpact questons would smply be another measure of efforts. To determne f ths s a cause for concern, we examne whether the responses to the questons about the effectveness of specfc actvtes are systematcally related to the responses to other questons n the survey, responses that we would not expect to be a consequence of a person s proenvronment behavors. In partcular, we estmate how people answer questons about the lkelhood that clmate change wll affect them personally and ther opnon on whether we worry too much about envronmental problems and not enough about prces and jobs. We estmate models that nclude both perceved mpacts and the frequency of pro-envronment behavors. 25 We fnd that the hgher the mpact on carbon emssons that ndvduals assgn to a gven actvty the more lkely t s that ndvduals beleve clmate change wll affect them and the more lkely t s that they dsagree that we worry too much about the envronment. For example, after controllng for actual recyclng efforts, we fnd that ndvduals who ncorrectly beleve recyclng half of a household s waste reduces carbon emssons by more than 2,500 pounds are more lkely to strongly dsagree we worry too much about the envronment (coeffcent sgnfcant at the 1 25 We estmate ordered probt models that also nclude ncome, educaton, gender, race, age, regon, and whether the respondents consder themselves envronmentalsts. 22

25 percent level). Smlarly, after controllng for efforts to purchase envronment-frendly products, ndvduals who beleve usng compact fluorescent lght bulbs has a hgh mpact are more lkely to beleve clmate change wll mpact them personally (also sgnfcant at the 1 percent level). Because these estmatons also nclude the actual pro-envronment behavors, these results suggest that the belefs about mpacts have addtonal explanatory power and that people s responses to the questons about perceved effectveness are not smply another measure of behavor. Rather, these responses seem consstent wth a person s overall vew of the severty and mportance of envronmental problems. Comparng fatalsts and non-fatalsts mght also shed lght on whether perceved effectveness nfluences pro-envronment behavors or f t s the case that people who do certan behavor smply assgn a greater effectveness to that behavor. Fatalsts thnk ther ndvdual contrbutons do not help to mprove envronmental qualty. Thus, ther belefs about the effects of several actvtes on carbon emssons should not be a sgnfcant factor n ther decson to contrbute (we would expect that altrusm and envronmental atttudes could affect ther efforts). On the other hand, f belefs about mpacts merely reflect efforts, then we should stll observe the same correlatons between perceved mpacts and efforts for ths group of ndvduals as well. Table 5 presents the results when we splt our sample nto fatalsts and non-fatalsts. As expected, belefs about the effectveness of dfferent actvtes do not correlate wth the frequency of pro-envronment behavors among fatalsts, wth sx of the eght coeffcents n the fatalsts regressons enterng nsgnfcantly. The exceptons are the belef that usng cold water has a hgh mpact and, margnally at the 10 percent level, the belef that usng compact fluorescent lght bulbs has medum mpact. In contrast, all eght coeffcents for the non-fatalsts enter sgnfcantly wth the expected sgns. The fact that the postve correlaton between 23

26 perceved mpacts and efforts does not generally hold for fatalsts suggests that t s not the case that ndvduals smply gve a hgher mpact to the behavors they undertake more often. What then are the factors that explan efforts by fatalsts? We fnd that a person s level of cvc-mndedness correlates wth ENERGY and PRODUCT (margnally wth FOOD) among fatalsts. In addton, self-proclamed envronmentalsts who are also fatalsts are stll more lkely to engage n pro-envronment behavors wth greater frequency than those ndvduals who do not consder themselves envronmentalsts. 26 In fact, the margnal effect of beng at least a weak envronmentalst appears to be stronger for fatalsts than for non-fatalsts. These results suggest that t s the combnaton of beng a fatalst and not an envronmentalst that has large negatve effects on the provson of the publc good. 5.3 Robustness Checks Whle our dscusson has focused on the results of ordered probt models, we also estmate ordered logt models and fnd smlar conclusons. A lmtaton of ordered models s the assumpton of parallel regresson. 27 When we estmate multnomal probt models that relax the assumpton of parallel regresson we fnd that we can draw the same nferences about the hypotheses of nterest. We also estmate OLS models treatng the behavors as numercal varables and fnd that the models explan between 23 and 26 percent of the varablty n the dependent varables. Gven that we use ndvdual level data, the goodness-of-ft of the models s qute good. Qualtatvely and n terms of statstcal sgnfcance the results are almost dentcal. Overall, the man results are very robust to dfferent estmaton methods. 26 Ths result s consstent wth fndngs by Kahn (2007). 27 We perform a Brant test after runnng the ordered logt models and fnd volatons of the assumpton for a few varables n each model. Importantly, for the perceved mpact questons, we only reject the assumpton of parallel regresson for LIGHTKNOW3 n the PRODUCT model (at the 5 percent level). 24

27 Fnally, we note that our model assumes ndvduals engage n pro-envronment behavor n order to receve a warm glow and the survey questons prompt respondents to report behavors that are done out of concern for the envronment. Even so, t s possble that ndvduals may be accurately reportng behavor, but stll engagng n some of these behavors to reduce household expenses, rather than to receve the warm glow. To valdate that our results are robust to the ncluson of varables related to ndvdual costs, we explot varablty across states n the opportunty cost of engagng n the behavors. Frst, we use a fner regonal categorzaton and estmate the models wth eght dummy varables for New England, Md-Atlantc states, East- North Central states, South-Atlantc, East-South Central, West-South Central, and Mountan states. Second, we use three state-level varables that may be related to ndvdual costs and benefts: the proporton of ndvduals n the state wth access to curbsde recyclng programs n the emprcal model estmatng frequency of recyclng, retal resdental electrcty prces (2006 averages) n the emprcal models for ENERGY and PRODUCT, and average prces farmers receve for beef cattle at the state level as a possble control n the model predctng FOOD. 28 The coeffcent on access to curbsde recyclng programs s postve and sgnfcant at the 5 percent level. The dummy varables for the perceved mpact of recyclng are stll postve and the dummy for hgh mpact s now statstcally sgnfcant at the 10 percent level. The coeffcent estmates on retal prces and prce receved for cattle are nsgnfcant. Importantly, the ndcators for perceved effectveness mantan ther levels of statstcal sgnfcance, provdng support for the clam that costs and benefts that accrue to the ndvdual are not the sole reason for engagng n the behavors. 28 For access to curbsde programs, we use 2000 data from the 12 th annual Bocycle natonwde survey (Bocycle magazne, Aprl 2000). We obtan 2006 average resdental retal electrcty prces from the Energy Informaton Admnstraton ( The prces receved of beef cattle come from the USDA Natonal Agrcultural Statstcs Servce. The results are avalable from the authors upon request. 25