Health Behavior Changes Through TeleGsion: The Roles ofDe Facto and.Motivated Selection Processes*

BRIAN R. FLAYUniversity of Nlinois ut Chicago

STEPHANIE McFALLUniversity of Oklahoma

DEE BURTONUniversity of Illinois ut Chicago

THOMAS D. COOKNorthwestern University

RICHARD B. tiARNECKEUniversity of Illinois at Chicago

Journal of Health and Social Behavior 1993, Vol. 34 (Decembcr):322-335

We assess the role played by two types of self-selecrion in accounring for theinfluence of a television series on smoking cessation. De facto selection is based onrespondents’ regular channel viewing habits that can expose them to the series.Motivated self-selection takes place when viewers deliberately select to watchtelevision programming because it meets their desire to quit smoking. Self-selectionalso can ,be viewed as a methodological artifact, spuriously accounting for theassociation between the airing of the series and smoking cessation among thetarget audience; Subjects were a probability sample of Chicago smokers whoregularly watch the evening news on one of the network channels. The interventionwas a televised self-help smoking cessation program broadcasr on one of thenetwork channels. over 20 days. Using nested covariance structure models for theanalysii, tie’tionclude that 1) de facto selection had no influence on exposure to theprogram’: 2) motivated selection had nd influence on exposure to the program: 3)the prograin reduced smoking; and 4) this effect cannot be attributed solely to themethodological artifact of self-selection, although motivation to quit smoking didhave the strongest influence on attempts to quit.I . :.. _I i... .<.;I ‘.*.. Y - --,

;~~:~.;~~~~~~~~~~~‘~~~~ :,ti ‘- ‘, :-:‘..?+: It I& pro&-$fficult to design mass media

:i,-t:heth promoti&’ campaigns that bring aboutonly 4 to 5 percent of smokers to quit and stayquit for a year or more, although supplement-

meaningful behavioi,change in a large propor-_. ._. .... tidti’ of exposed persons. For example, tele-

ing these clinics with written materials may dou-

.lvised smoking cessation clinics .typicaIly helpble effectiveness (Flay 1987a, 1987b). Media

..-.. programs have low success rates for many rea-_,_r .,,.I._ ..r .“$‘,.. ,, .;. _‘_. sons, including not reaching the target audience

.i-3.‘. 7: 9:. :. -i:,r,._. . . . . :,. ..I*-*.. ,a;:. .*:r., ..I&: 1. ‘9. T,. ,. enough times, not gaining the attention of theI .- ,.; :;,,t..,,. ::. . .

-..;‘,t Dimct a!l.;correspondence to Brian R: Flay, target audience even when they are reached,FY&nti& Research Center, University of Illinois using messages that are not fully comprehensi-

-‘.:. at Chicago, Chicago,. IL 60607-3025. This work ble to the target audience, and using ineffectivewas supported by National Cancer Institute grant messages (Flay and Burton 1990). In this study,

nd. CA42?60., We,thank Ohid Siddiqui and Steve we assess how two different types of self-,. Serio for theii help wit&the data analysis. selection (Sears and Freedman 1967) might ac-

322 -

HEALTH BEHAVIOR CHANGES THROUGH TELEVISION 323

count for a television series’ influence on smok-ing cessation.

Sears and Freedman (1967) suggested thatself-selection may be de facto or motivated.De facto selection depends on respondents’regular channel viewing habits, which willonly expose them to a health promotion seriesif it is broadcast on the channel they prefer.Motivated self-selection operates when view-ers deliberately select to watch a televisionprogram because it meets a specific need-inthis study, to stop smoking., Self-selectionalso can be a methodological artifact, spuri-ously accounting for an association betweenthe airing of a series and smoking cessationamong the target audience.

De Facto Selectivity

Earlier experiemental social psychologicalresearch concluded that de facto selectivity iswidespread and that individuals acquire muchof what they learn as a product of exposuresthey have not sought out (Freedman and Sears

1965; McGuire 1968, 1985: Sears 1968).This being the case, it is no surprise that somescholars have called for studies of thedeterminants and consequences of de factoselectivity (Cotton 1985; Katz 1968; Klapper1949, 1960; Sweeney and Gruber 1984;Zillman and Bryant 1985). For instance,women read more health-related stories thanmen, probably because these particular storiesappear more often in magazines read mostlyby women for other purposes (Feldman1966). People not employed full-time andolder persons are more likely than others tosee almost any program on television simplybecause they watch more television thanyounger or employed persons (Real, Ander-son, and Harrington 1980; Smith and Moschis1984). Can changes in health behavior (e.g.,smoking cessation) result from incidentalexposure to media outlets, especially whenthere is no prior desire to change? We refer tothis as the “de facto selectivity” hypothesis(represented by path A in Figure 1).

Why is the de facto selectivity hypothesisimportant? If viewing as a result of de facto

Figure 1. Theoretical Model of Selection ProcessesREGULARV I E W I N G \HABITS A

WATCH B TARGETh PROGRAM BEHAVIOR

“’ PRIORMOTIVATION

FORBEHAVIOR

CHANGEC

,a

D

I ’

CHANGE

..SELECTION PROCESSESI

De facto selection = AMotivated behavior change = D

> Motivated selection = C

324 JOURNAL OF HEALTH AND SOCIAL BEHAVIOR

selection does indeed lead to higher rates ofsmoking cessation, then campaign effective-ness depends on successful dissemination-reaching the target audience in large num-bers-rather than on any strategy that wouldinvolve targeting individuals already willingto change. Moreover, in most areas of thecountry, media professionals know the demo-graphics of the audience usually reached byany particular program. Hence, if one wantedto reach a sub-audience with a particularprofile-in our case, city residents who arepoor and of color-it would be possible toselect media outlets that disproportionatelyreach them. Reaching large numbers ofsmokers who do not want to quit neverthelessmay cause some of them to quit because oftheir exposure to televised smoking cessationmaterials that they had not been particularlymotivated to seek out for themselves.

Motivated Selection

The motivated selection hypothesis postu-lates that people who are motivated to make aparticular change will actively seek outprograms to help them make that change.Katz (1980) suggests that such selectivity is amajor mediator of mass communicationeffects because it influences both the nature ofthe content to which someone is exposed andhow deeply he or she processes the message.From this perspective, it seems reasonable tohypothesize that people who are more ready(motivated) to stop smoking will be morelikely than others to seek out and view atelevised smoking cessation program and toprocess the program content more deeply inways that add to program effects over andbeyond those that might be attributed tomotivation alone. We refer to this as the“active motivated selective exposure*’ hy-pothesis (represented by path C in Figure 1).As obvious as this hypothesis seems, it hasnot been corroborated directly. Indeed, earlierexperimental social psychological researchfound little evidence for widespread selectiveexposure to materials with a particular pointof view (Freedman and Sears 1965; McGuire1968, 1985; Sears 1968). and some scholarshave suggested that a lack of motivatedselective exposure to media programs is onepossible reason for a lack of media effects(e.g., Klapper 1949, 1960; Roberts andMaccoby 1985).

Self-Selection as a Methodological Artifact

A different type of motivational hypothesissuggests that people motivated to changeoften do so independently of exposure to atelevised program on the topic. If programeffects are correlated with such motivation,then a self-selection confound can result, withanalysts attributing behavior change to theprogram rather than preexisting motivation.Such a pattern is often viewed as a methodo-logical artifact without substantive value forexplaining media effects (Sears and Freedman1967). We refer to this as the “motivatedbehavior change” hypothesis (represented bypath D in Figure 1). Before any effects can beattributed to a televised program, we mustrule out the possibility that any relationshipbetween the program and changes in smokingrates is not due to path D-that is, thosepeople who change happen to be those whowould have changed anyway. ’

Study Hypotheses

This paper presents an empirical assess-ment of the relative importance of the aboveconceptualizations of self-selection in deter-mining the effects of a televised smokingcessation program. Literature in communica-tions research and experimental social psy-chology predicts that exposure to the televisedsmoking cessation program will be deter-mined more by general media use habits (pathA in Figure 1) than by motivation to quitsmoking (path C) (Gerbner et al. 1986;Snyder and Ickes 1985). especially as theprogram was promoted only on the channelon which it was delivered. However, wewould not expect path C to be insignificant.All rational models of human behaviorpostulate some form of “active motivationalselection’* whereby individuals seek out theinformation they need to pursue goals theyhave set for themselves-including the goalof quitting smoking. We also expectedmotivation to quit to predict smoking cessa-tion (path D).

The major intervention hypothesis is that ex-posure to the television program would influ-ence quitting (path B), indicating that the im-pact of the televised program was greater thanif it had not occurred at all. A significant pathB would strongly suggest that motivation tochange cannot explain all of the observed ef-

HEALTH BEHAVIOR CHANGES THROUGH TELEVISION 325

fects on smoking cessation, since somethingother than motivated behavior change (the meth-odologist’s self-selection confound) would beresponsible for at least some of the observedreduction in smoking. A significant path B wouldsuggest effects of de facto and/or motivated self-selection.

Several demographic variables were in-cluded in our analyses as potential predictorsof viewing and motivation. First, based ondata from the television station, we expectedthat socioeconomic status would predict newsviewing. In particular, the Channel 7 audi-ence at that time was more African-American, more blue-collar, and less edu-cated than the population of Chicago newsviewers as a whole. We confirmed this withbaseline data collected six months before theintervention. Data from other research onattention to health news suggest that females,particularly those who have less than a highschool education or who are African-American, are most likely to obtain healthinformation from televised news (Manfredi etal. 1984). Hence, we include gender in ouranalyses. Other data suggest that thoseemployed full-time and older persons watchthe news more often than others (Real,Anderson, and Hanington 1980; Smith andMoschis 1983; Warnecke et al. 1991),leading us to include these variables in ouranalyses. Finally, we expected some of thesedemographic variables to predict motivationto quit smoking as well as to predict viewing.Motivation to quit smoking is usually loweramong the less educated, females, African-Americans (Manfredi et al. 1990). andheavier smokers (Helman et al. 1991).

METHODS

The Intervention

The intervention was a televised self-helpsmoking cessation program based on theAmerican Lung Association’s “FreedomFrom Smoking in 20 Days” (see Wamecke etal. 1991 for a detailed description). Theprogram was broadcast during the localevening news at 4 p.m. and 10 p.m. almostevery day for three weeks during February,1987, on Channel 7, WLS-TV, the ABCnetwork-owned station in Chicago.

Smokers could obtain the American LungAssociation (ALA) manual that accompanied

the televised series directly from participatinghardware stores or offices of the participatingHMO. Alternatively, they could obtain amanual by first-class mail if they registered ata participating hardware store or (during thefinal days just before the series started) calledan “800” number. During the three-weekregistration period, lo- and 30-second spotsfeaturing a mime “kicking the habit” werebroadcast (only via Channel 7) which told‘smokers how to register.

The overall project included studies ofprogram participants, examining such issuesas the incremental effects of various programcomponents (Wamecke et al. 1991). Theanalyses reported here are on a panel ofsmokers drawn for comparative purposes’from the general population of smokerswithin the six-county Chicago MetropolitanStatistical Area (MSA).

Sampling and Data Collection

The sample was selected using randomdigit dialing procedures. Approximately29,000 households were screened for smok-ing status and news viewing. For eachhousehold, all smokers were listed, and onesmoker was randomly selected for an inter-view. Those smokers who djd not view newson television (approximately 13%) were notfollowed.* Viewers of Channel 7 news wereoversampled. The final panel was composedof roughly equal numbers of viewers from thetarget and non-target channels (resultingsample weights of 1.16 and .80, respec-tively). From the screening, 3,299 eligiblerespondents were identified and 2,398 wereinterviewed (Time 1 [Tl]). About 85 percentof these were reinterviewed at the secondpretest (Time 2 [T2]) and 1,560 (65%) werereinterviewed at the immediate posttest (Time3 [T3]).3 The primary reason for attrition wasfailure to reach the respondent because ofunlisted numbers when the respondent moved(Warnecke et al. 1992).4 Of the 1,560respondents available at the’ second posttest(Time 4 [T4]), 200 were excluded becausethey had quit smoking by T2 and thus werenot asked about their plans to stop smoking.In addition, 102 cases were deleted becauseof missing income data, 22 cases because ofmissing responses on indicators of motivationto stop smoking, and 28 cases because ofmissing values on one of the other modelvariables.

326 JOURNAL OF HEALTH AND SOCIAL BEHAVIOR

Table 1 indicates the characteristics of thesample at Tl and T3 (end of the program),and the composition of the analysis sample.According to Table 1, while attrition washigher among males, among the young, theless educated, and those with lower incomes,these differences were minor.

TABLE 1. Characteristics of the AnalysisSample and the Complete Samplesat Tl and T3

Characteristic

Tl T3 AnalysisSample Sample Sample

(96) (96) (%IEducation

Less than High SchoolHigh School GraduateCollege Plus

IncomeLess than or equal

to $13,000$13.001-$25,000$25.001-640.000More than $40.000

Employment StatusFull-time ..Part-timeNot Employed

Race (Black)Gender (Female)AE

173031-45

65 and aboveFrequency of Viewing

10 p.m. Channel 7 Newsfdays per week10 days1-2 days3-4 days5-7 days

Watch Any InterventionYes

Smoking Status at T2Nor smoking

If smoking, number Of:* cigarettes smoked daily

‘; Less than 20.” . 20-29

3cL3940 or more

Motivation (Plan to Quit):. Never. . In more than 12 months

Within 12 monthsWithin 6 months

Within 3 months;j Stop Smokine 24 hours

..ys ,

13.1 11.5 11.036.0 35.3 36.450.9 53.2 52.6

14.7 12.8 12.123.8 23.1 24.531.1 31.7 32.130.4 32.4 31.3

56.7 56.2 57.413.4 14.5 13.730.0 29.4 28.916.9 16.0 15.655.3 56.8 58.3

28.2 26.7 26.135.4 35.3 36.427.7 29.5 29.68.7 8.5 7.9

43.3 43.7 44.017.6 18.7 18.418.2 17.2 17.620.9 20.3 20.0

26.1 27.6

12.0 12.8 N/A

N/A 42.0N/A 35.8NIA 11.7N/A 10.4

. 9.847.212.112.818.1

45.3

41.636.011.810.8

10.940.014.515.219.4

38.7

. . Nores: Number of subjects varies because of missingitems: Tl, n = 2.201-2.398; T3. n = 1.444-1.560;

‘.’ Analysis Sample, n = 1.208.

Approximately one-half of the respondentshad education beyond high school, with morethan one-half having family incomes greaterthan $25,000. Employed persons representedmore than 70 percent of respondents. Thesample had a substantial representation ofBlacks (approximately 16%) and of lowerincome persons. The representation of malesand older persons (over 65 years) of age wassomewhat lower than their representation inthe Chicago MSA, reflecting lower participa-tion by males and lower rates of smoking bythe elderly. Slightly less than 60 percent ofthe sample reported watching the news at 10p.m. on Channel 7 in the week before theinterview. The 10 p.m. news is the mostpopular time for local news viewing inChicago. Approximately 20 percent of thesample watched Channel 7 news regularly(five to seven days in that week). Over 25percent saw at least one segment of theintervention. Twelve percent of the respon-dents reported not smoking at the secondpretest. Of those reporting smoking in thisinterview, approximately 40 percent smokedless than a pack a day (assuming 20 cigarettesin the average package), and 10 ‘percentsmoked more than two packs. This pattern ofsmoking is similar to that found in theNational Health Interview Survey (Novotny etal. 1989). but lower than reported byparticipants in the smoking cessation program(Warnecke et al. 1991). Almost 20 percent ofthe smokers planned to quit smoking withinthree months of the T2 interview. About 40percent reported having quit for 24 hoursduring or after the intervention.

Measurement Model

We classified the variables used in the modelas representing so&demographic and controlvariables, news viewing on the target channel,motivation to quit smoking, watching the inter-vention, and attempts to change behavior (rep-resented by stopping smoking for 24 hours).

Sociodemographic and Control Variables.These variables include education, gender,race/ethnic background (Black vs. other), andemployment status (full-time, part-time, andnot employed). In addition, we controlled forthe average number of cigarettes smoked perday, measured at T2.

News Viewing. The frequency of Channel 7

HEALTH BEHAVIOR CHANGES THROUGH TELEVISION 327

news viewing was assessed by two indicators:the number of times Channel 7 news waswatched during the previous week at 4 p.m.and at 10 p.m. These indicators weremeasured at T2.

Motivation. The motivation to quit smok-ing was assessed by subjects’ reports of plansto quit smoking. Those subjects who hadplans to quit were asked when they planned tostop smoking. Responses were preceded toinclude: in more than 12 months, in less than12 months, in less than six months, or in lessthan three months. Those with no plans toquit were coded “never.”

Watched Intervention. Subjects were askedpost-intervention whether they watched the“Freedom From Smoking in 20 Days” TVprograms and the number of days per weekthey had watched. The measure of exposureused in these models was whether the subjecthad watched any of the programs.s

Behavioral Outcome. The outcome exam-ined was whether the individual stoppedsmoking for at least 24 hours following theintervention. This measure includes respon-dents abstinent at T3 (8.9%) and those whostopped smoking for at least 24 hours betweenthe end of the program and T3 (29.8%).

Plan of Analysis

We used covariance structural analysis inthis paper. Since most of the observedvariables were ordinal, or categorical, we usedversion 1.20 of PRELIS (Joreskog andS&born 1988b) to compute the polychoric/polyserial correlations between all pairs ofvariables in the models. Maximum likelihoodestimates of the models were obtained withversion 7.20 of LISREL (Jbreskog andS&born 1988a).

Although driven by theory, fitting struc-tural models proceeds iteratively. One speci-fies and estimates a model, examines theresults and modifies it, adding or removingparameters or constraints to improve the fit toone’s sample of data. Such specificationsearches can produce invalid and unstablemodels that must be viewed with caution(MacCallum 1986). To protect against sam-ple-specific errors, we adopted a cross-replication scheme.6 We randomly dividedour cases into a development sample onwhich to explore the data and refine our initialmodel specification,’ and ‘a test sample on

which to re-estimate parameters using thesame model specifications. We required theratio of the parameter to its standard error toexceed 1.96 in magnitude in both subsamplesfor replication. When the replication criterionwas not met, we nonetheless examinedpatterns in the parameters. In the majority ofcases in which parameters did not meetcross-validation criteria, the specific parame-ters were in the same direction but statisticallysignificant in only one of the subsamples.Results are presented for the total sample, andthe development and test samples separately.Parameters that meet the cross-validationcriteria are indicated.*

The general model estimated is shown inFigure 1. For simplicity, we omit details ofthe measurement model and of the errorstructure. Also, we do not show the pathsfrom the sociodemographic variables to theendogenous variables. The path labelled Asummarizes the effect of Channel 7 newsviewing on watching the intervention. Path Bis the effect of watching the intervention onstopping smoking. The effects of motivationare represented by paths D, the direct effectof motivation on stopping smoking, and C,the effect of motivation on watching theintervention program.

RESULTS

Analysis Strategy

Models representing different aspects ofselection were estimated and compared. Table2 summarizes the fit of five nested modelsand the final selected model. One model isnested within a second when its free parame-ters are a subset of those in the second model.Each of the first five models are successivelynested. The selected model represents a minorvariation that simplifies the structure of themodel by constraining one of the pathsidentified as nonsignificant. This final modelis not nested within its preceding model. -::

In nested models, the effects of adding suc-cessive sets of parameters can be assessed viathe increment in &i-square. The sequence inwhich parameters or sets of parameters are addedand assessed is theoretically ordered. In Model1, the so&demographic variables and level ofsmoking influence media habits, motivation, pm-gram exposure, and attempts to quit smoking.This model serves as our baseline for cornpat+

328 JOURNAL OF HEALTH AND SOCIAL BEHAVIOR

TABLE 2. Goodness of Fit for Nested Structural Models of Selection ProcessAdjusted

Chi-square Goodness Goodness Incremental Chi-Description of Model (df) of Fit Index of Fit Index square (df)

Model 1 395.53 .95 .84Demographics Only (22)

Model 2 320.46 .96 .86 75.07**+Add Motivation to (21) (1)Quit SmokingPaths from Figure I: D

Model 3 320.34 .96 .86 .I2Add Motivation to (20) (1)

Watch InterventionPaths: D, C

Model 4 312.75 .96 .85 7.59**Add Watch Intervention 09) (1)to Quit SmokingPaths: D, C, B

Model 5 112.70 .98 .94 200.05*** ,Add News Viewing to (18) (1)Watch InterventionPaths: D. C. B. A

Model 6 (not nested) 107.64 .98 .94Trims Model Constraints (19)Deletes Motivaton toWatch Intervent ionPaths: D, B, A

fores: P&ability for incremental chi-square: ** p < .Ol; *** p < .OOl.

sons and is a minimal defensible model as dis-cussed by Sobel and Bohrnstedt (1985). The spe-cifically hypothesized paths from’ theso&demographic variables to endogenous vari-ables are shown in Table 3. Model 2 retains theeffects of the so&demographic variables as inModel 1, but adds path D, representing the di-rect effect of motivation on attempts to quit smok-ing. Model 3 adds a path from motivation towatching the intervention. It corresponds to theeffect of motivated selection. Model 4 adds thepath from watching the intervention to stoppingsmoking. It tests the effects of the program onbehavior. Model 5 adds a path from Channel 7news viewing to watching the intervention. Thisis a test of de facto selection effects. Model 6makes a minor modification by trimming an un-substantiated relationship.

A comparison tif Models 1 and 2 supportsthe importance of motivation in behaviorchange. However, the relationship betweenmotivation and watching the intervention wasnonsignificant, as shown in the comparison ofModels 2 and 3. Thus, the role of motivatedselection was not supported. The addition ofthe path from watching the intervention tostopping smoking made a significant contri-bution, providing some support for the

. importance of program exposure for short-

term behavioral change. The comparison ofModels 4 and 5 shows strong support for theimportance of de facto selection.

Since the component of motivated selectiondid not contribute to the overall model, we setthat parameter to zero and estimated Model 6for detailed presentation, While Model 6 is notnested within Model 5, its goodness of fit re-sidual chi-square was equivalent to that of Model5 (107.64 with 19 degrees of freedom[p<.OOl]). The Goodness of Fit Index (GFI)was .98 and the adjusted GFI was .95. We alsopresent the fit on identical models estimated onthe development and test samples. The devel-opment sample had a chi-square of 68.05, with19 degrees of freedom (p<.OOl). Its GFI was.98, with adjusted index of fit of .94. The chi-square on the test sample was 77.92, with 19degrees of freedom (p<.OOl), a GFI of .98,and adjusted index of fit of .92. The fit of Model6 is satisfactory.

Table 3 presents parameter estimates andstandard errors for Model 6.9 The results aresummarized for the total sample, and thedevelopment and test samples. Paths that metcross-validation criteria are indicated. Figure2 shows the parameter estimates for the majortheoretical paths from Figure 1.

HEALTH BEHAVIOR CHANGES THROUGH TELEVISION 329TABLE 3. Unstandardized Parameter Estimates and Standard Errors for Selection Model 6

Channel 7 Motivation Watched Stop forPrecursor News Viewing To Quit Intervention 24 Hours

Variables Sample Coef. S.E. Coef. S.E. Coef. S.E. Coef. SE.

Education Total - .055* (.023) .047 (.028) .080’ (.032)Develop - . I 18* (.035) .048 (ml) .ll8* WWTest -.007 (.028) A48 (.@w .029 t.044

Employ Total .204*c (.030) -.080 C.046)Develop .183* COW - .205* LOWTest .225+ (.038) - .034 (.067)

Race (1s Black) Total .112+ (020) .059* (.029)Develop - .OO 1 (028) 061 w42)Test .184* (028) .077 um

Age . Total - .017 (024) - .09O*C (.028) .2OO*c (.033)Develop - .009 (.036) - .081* wm .199* (.045)Test - .029 (030) - .099* (.OW .214* (.047)

Gender (1 = Male) Total .024 (.026) .003 (.029) -.123* (.035) . l45*C (026)Develop .033 (.042) - .047 WI) - .282* (053) .080* (.039)Test .013 (.030) .046 (.041) - .oOl (.047) .207* (.034)

Amount Smoked Total - .172*C (030) - .415*C (.026)Develop - .157* (.042) - .363* (.039)Test -. 182* (.043) - .466* (.034)

Cb 7 News Viewing Total 1.248*C (. 135)Develop 1.226* (188)Test I .392* (.202)

Motivation Total .228*C (025)Develop .165* (.038)Test .283* (.034)

Watched Interven. Total .058* (025)Develop .088* (038)Test .024 (.033)

R* Total .I99 .052 .489 .258Develop .158 .050 .516 .I80Test .360 .061 .488 -356

C = cross-validation criterion met.* = p < .05.

Predicting Regular Viewing Habits

News viewing on Channel 7 was higheramong the employed. A number of other rela-tionships were significant in the total sample,but failed to replicate in the cross-validation.Within the total sample, Blacks and individualswith less education watched the news more fre-quently on Channel 7, but this was not repli-cated in the development sample. There wereno relationships with age and gender.

Predicting Motivation to Quit

Motivation to quit was higher amongyounger smokers and those who smokedfewer cigarettes per day. Blacks reponedhigher motivation in the total sample but thisrelationship was not statistically significant ineither of the subsamples. Blacks generallysmoked fewer cigarettes than Whites, al-

though their smoking was more concentratedin brands with high tar and nicotine (Novotnyet al. 1989). Relationships between educationand motivation were not observed.

Predicting Exposure to the Intervention

In both samples, the frequency with whichpersons regularly watched the Channel 7news was the primary determinant of havingwatched any of the smoking cessation inter-vention, This supports the de facto selectionhypotheses. Both samples also showed thatolder respondents watched the interventionmore often than others. In the total sample,women in general and people with moreeducation were more likely to watch theintervention, but these relationships were notreplicated across the samples. No relationshipwas observed with employment status.

.

330 JOURNAL OF HEALTH AND SOClAL BEHAVIOR

’ Fii 2. Path Coeficients for the Fii3 Fitted ModelREGULARVIEWINGHABITS

A= 1.248

.\ .WATCH B=O.O58 TARGETw PROGRAM- BEHAVIOR.

CHANGE

CHANGE

SELECTION PROCESSES

De facto sektion = AMotivakd behavior change = D

Motivated selection = C

Predicting Attempts to Quit Smoking one of the subsamples. Since the 95 percentconfidence intervals for parameters in the two

Motivation to quit and its determinants subsamples overlap, there is some evidencewere the strongest predictors of attempts to of a small effect of watching the program andstop smoking. Males, those who were more quitting smoking over the short term. Notemotivated to quit smoking, and light smokers that this effect is produced in a model whichwere more likely to stop smoking for at least includes the motivation to quit smoking and24 hours following the intervention or to be various demographic factors. These serve asabstinent. For example, those Channel 7 news part of the model that takes statistical accountviewing smokers who planned to quit within of selection as a methodological artifact. Ourthree months (the most highly motivated) conclusion about a small treatment effect is ofwere twice as likely (12.6%) as those who course contingent on its adequacy as anplanned to quit within a year (6.5%) to report appropriate model of self-selection and be-trying to quit with the program; these people havior change.wem, in turn twice as likely to try to quit withthe program as those who did not plan to quitwithin a year (3.2%). DISCUSSION

Eflects of Program Exposure_. _.

Our suggested model of the role ofself-selection was partially supported. Weobtained clear support for the view that de

: Although the effect of program exposure facto self-selection influences whether audi-on. attempts to quit smoking (path B in:: ences view particular media programs. and

Figures 1 and 2) was not strong, it was that this viewing then influences behaviorstatistically significant in the total sample and change. We found no support, however, for

: .:,

:_,.=..~.5.. .,. .I ;..)..:.‘.

-:

.,-

HEALTHBEHAVIOR CHANGES THROUGH TELEVISION 331

the motivated selectivity hypothesis in thissample. Smokers do not seem to haveswitched channels to watch the cessationprogram in the form in which it was promotedin this study.

Consistent with part of the de factoselectivity hypothesis was the fact thathabitual Channel 7 viewing was the bestpredictor of exposure to the smoking cessa-tion program (see path A in Figure 1). Thefact that exposure to the programmingincreased 24-hour quitting after a variety ofmotivational and demographic factors weretaken into account also was consistent withthe second part of the hypothesis. Takentogether, these paths suggest that particularaudiences can be successfully targeted andsome change brought about merely bydetermining which group views a particulartelevision channel most often and knowingthat the televised content meets high substan-tive standards. In this case, the effort wasdirected at a wide audience in the Chicagoarea, but special emphasis was placed on poorfamilies and African-Americans whose smok-ing rates are particularly high and whose

attendance at smoking cessation clinics issporadic. Broadcasting the intervention on achannel known to attract households withsuch a profile did indeed gain an audiencethat was disproportionately poor and of color.Yet all of this was accomplished within aframework in which literally hundreds ofthousands of persons were exposed to thetelevised smoking cessation materials.

Consistent with the motivated behaviorchange (and methodological self-selection)hypothesis, the strongest predictor of attempt-ing to quit was prior motivation to quit. Butthe motivated selectivity hypothesis was notsupported in so far as motivation to quit didnot predict viewing of the smoking cessationprogram. Prior motivation to quit does notseem to have been very successful in gettingsmokers who wanted to quit to switch overfrom one channel to another. To the extentthat loyal viewers on another channel wereinformed about programs airing on Channel7, channel loyalties overcame any desire to beexposed to anti-smoking programming. Chan-nel 7 did advertise its smoking cessationprogram in the press and its own program-ming, but it obviously did not try to buy timeon a competitor’s channel to advertise there!

These findings confirm the observations ofrirany others regarding the difficulties of using

mass media for health promotion (Flay et al.1988; McGuire 1986). On the one hand, thosemost likely to view the program normallyview the channel at the time of the program.These people are not likely to be especiallymotivated to change. On the other hand, thosepeople most likely to change are already mostmotivated to change. Since those most likelyto change are not motivated to seek out healthpromotion programs on media channels theydo not usually use-at least not with thelevels of advertising of the intervention usedin this study-the media programming haslimited success with them.

Fortunately, our other findings suggestways around this dilemma. First, viewing theprogram was a significant predictor ofquitting smoking for the complete sample,and in each subsample the coefficients had thesame sign and overlapping confidence inter-vals. The implication is that de facto selectionwill facilitate quitting because it gets somepeople to view who are influenced by whatthey see even if they were not originallymotivated to change their behavior. Thisfinding is consistent with the suggestion byKlapper (1960) and Cotton (1985) that defacto selectivity might play an important rolein explaining media effects.

At a much more speculative level, analternative view of motivated selectivitywould argue that readiness for change is morepassive than going out of one’s way todiscover relevant media materials that willsupport a particular change. Persons “some-what” motivated to change their behaviormay discover a program that will help themmake the desired change, even though theirmotivation to change is not strong enough toseek out the program. Two health behavior/promotion theories adopt this view, positingthat a program can become a “cue to action”(the Health Belief Model; Becker 1974) or an“enabling factor” (the PRECEDE model;Green et al. 1980), especially for those whoare not opposed to change or Completelyindifferent to it. Some communications re-searchers also hold this view, considering thatself-selection is attributable more to situa-tional than motivational factors (Comstock1986; Cotton 1985). From this pnPeCtiV%people who are ready to change might notovertly seek out supportive information, butmight instead stumble upon it and when theydo, attend to it and then :pdCiPate inprogram-sponsored activities if .it iS COnVe-:

332 JOURNAL OF HEALTH ANIJ SUUAL DmAV WI\

nient to do so. They may not actively searchfor a program, however, or switch from theirfavorite channel to another. We refer to thisas the “cued motivated selective exposure”hypothesis. The data from this study do notallow for a test of this hypothesis becauseanother measure of motivation to changewould need to be collected after the initialviewing on the target channel.

An effective extended public health inter-vention must reach the greater numbers ofless motivated smokers among the targetaudience. Our data demonstrate quite clearlythat media programming of this type reachesmany people not already motivated to change.Of those Channel 7 news viewing smokersMore motivated to quit (that is, planning toquit within three months; n = 86), 45 percent(n = 39) saw the program; of those, 61 percent(n = 24) stopped smoking for at least 24 hoursfollowing the intervention. Of those Channel7 news viewing smokers less motivated toquit (n =368), the same proportion (45%,n = 165) saw the program and a lowerproportion (35%) but higher numbers (n = 59)of them stopped smoking for 24 hours. Thatis, more than twice as many less motivated as

more motivated smokers quit smoking withthis program. Though people motivated tochange may be more likely to register for aprogram, view it, and attempt to quit with it,greater numbers of less motivated people canstill be reached because of de facto selection.Further studies are needed to disentangle theextent to which these observed patterns aredue to direct effects of motivation to quit onactual quitting behavior (the methodologicalartifact in this context) or indirect effects ofmotivation through program participation(cueing). Regardless of the causal pathsinvolved, future program developers might domore to motivate the less motivated membersof the target audience to change beforepresenting them with a program designed toeffect that change.

Exposure to a mass media health behaviorintervention is largely determined by de factoselection-by habitual patterns of media userelated to the general level of televisionviewing, channel loyalties, and programpreferences. In turn, de facto selection isdetermined mostly by demographic variables,

although some psychosocial and personalityvariables are likely to be influential (Kviz etal. 1991). Active motivated selectivity ap-pears to play only a minor role in determiningexposure: in this instance, few personsswitched channels to view the offered pro-gram. More might have done so had therebeen even more publicity about the smokingcessation program in outlets read by peoplemotivated to change. But with the level ofpublicity achieved here, there is no evidenceof systematically changing channels to beexposed to the televised smoking cessationmaterials. Future research needs to examinethe role of cued motivated selective exposureto test whether exposure causes an increase inthe motivation to quit in at least some kinds ofpeople.

The findings that 1) channel preferencepredicted exposure and 2) exposure predictedquitting after controlling for some selectionfactors, suggest that this media programcaused a reduction in smoking that would nothave occurred during the same time periodwithout the program. In the future, research-ers will need to differentiate between differenttypes of selection processes. They also willneed to determine the portion of theirobserved effects due to the motivation tochange considered by methodologists to bespurious, and the various sources of selection(de facto, active motivational, and cuedmotivational) that can legitimately be consid-ered to be the causes of any changes producedby a media program. Examining these issueswill require collecting motivational data on allstudy participants prior to an intervention andat a later time after initial exposure but beforemultiple exposures and behavior change.

NOTES1. The purpose of the panel in the larger study was

to permit comparisons between program regis-trants and non-registrants.

2. So&demographic information was collectedon a subsample -of 500 smokers who do notwatch the local television news. Persons whowere older, female, Black, and had familyincomes above $13,000 were more likely thanothers to watch local television news at leastthree times per week.

3. Additional interviews were conducted sixmonths, one year, and two years following theintervention. The data reported in this paper arelimited to the two pretests (Tl and T2) and theimmediate posttest (T3).

HEALTH BEHAVIOR CHANGES THROUGH TELEVISION 333

4. Attrition can produce selection bias that coulddistort estimates of relationships among thevariables. We performed logistic regressionanalyses to predict attrition at the second andthird interviews from demographic variablesand television news viewing at the time of theinitial pretest (Tl). Attrition was higher amongyounger and lower income adults and males.Attrition was not related to baseline smokinglevel. We conducted some analyses to assessthe effects of attrition bias on exposure to theintervention or on attempts to quit smoking. Weconstructed a bias correction term using theresults of our logistic regression models forattrition. The correction term is a variable witha value for each case equal to the density of thelogistic distribution divided by the cumulativedensity at the predicted logit value for attritionfor the case. This bias term is used in the sameway as the Heckman estimator except that it isbased on a logistic rather than a probit model(Heckman 1979). Under certain conditions, thisfactor can be used to test and correct forselection bias due to systematic attrition. Weregressed the correction term on post-interven-tion and wave 2 variables, controlling for allmodel variables from Tl. The bias correctionterm did not have a significant relationship withany of the wave 2 (motivation measures) orwave 3 outcomes (attempt to quit smoking).

5. Alternative models using an indicator of howfrequently the respondent had watched theintervention (not at all, less than once a week,once a week, 2-6 days per week, and daily)obtained the same pattern of results.

6. The approach was similar to that used inresearch by Jersey Liang which was describedin a series of papers (e.g.. Liang et al. 1987).Because dividing the sample in half reduces itssize, this approach would not always beappropriate. Considerations would include howsmall the resultant halves would be and thestrength of anticipated relationships. Studiesfocusing on the measurement model canfrequently anticipate stronger relationships thanthose focusing on the structural model (as thisone). A large sample will permit the identifica-tion of a larger number of small effects, whichmay not be of practical importance. A samplesize of 600 is not generally regarded as small,and should permit identification of large tomoderate relationships. This study can thusserve as an application of this approach toanalyses focusing on structural relationships.

Alternative approaches are sample reusemethods like bootstrapping or the jackknifemethod. Although these approaches may be thebest ways of dealing with the stability ofestimates, they are in early stages of development for complex models such as covariance

structure. models. The situation we face herewould be the estimation of model parameters toobtain measures of variability. Efmn (1987)suggests 100 or more replications to estimatestandard errors of a parameter. Bollen and Sine(1990) point out that this is computationaI!ydemanding in LISREL or similar programs.They suggest reducing computational effort byusing previous values as start values andlimiting calculations to a single iteration.However, our problem also would involve 100or more sets of iterative calculations to producethe matrix of polyseriaI/polychoric correlations.

7. In general, the specification searches took thefollowing directions: a) attempts to model thestability of media use, including experimenta-tion with induced variables to represent mediause, b) deletion of indicators concerning thehelpfulness of a self-help manual and oftelevised smoking cessation programs in quit-ting smoking, c) deletion of a poorly distributedindicator measuring obtaining the ALA manual,and d) minor modification based on modifica-tion indices. Based on the modification indices,we permitted a correlation between measures ofnews viewing at 10 p.m. between the Tl andI?! waves. We subsequently simplified themodel to use only indicators from the ‘I2 wave.The most successful thrust of the specificationsearches, in that it resulted in more stablemodels, was in model simplification. As noted,we dropped indicators of news viewing fromthe first pretest. We also eliminated incomefrom the model to reduce collinearity with raceand education, and we deleted household size.Household size was originally influential whennews viewing was divided into early viewingand late viewing consttucts, but had noassociation with a combined news viewingconstruct. Larger households were less likely towatch the early news, probably because of mealpreparation and employment responsibilities.Finally, we conducted a search for multivariateoutliers as described by Bollen (1989). Thissearch identified as potentially problematicthose respondents with unusual combinations ofvariables: for example, an 80-year old respon-dent who smoked three to four cigarettes perday.

8. We investigated the gravity of failure toreplicate in the cross-validation process bycalculating 95 percent confidence intervalsaround the parameters in the test and develop-ment samples. With two exceptions, theconfidence intervals overlapped.

9. Parameters for the measurement model areavailable from the authors.

-

334 JOURNAL OF HEALIH AND SOCIAL BEHAVlOR

REFERENCES Heckman. lames J. 1979. “Sample Selection Bias

Becker, Marshall H. (cd.). 1974. The HealthBelief Model and Personal Health Behavior.

Thorofare, NJ: C.B. Slack, Inc.Bollen, Kenneth A. 1989. Structural Equations

with Latent Variables. New York: John Wileyand Sons.

as Specification Error.” Ecottometrica 46:153-161.

-and Robert Stine. 1990. “Direct andIndirect Effects: Classical and Bootstrap Esti-mates of Variability.” Pp. 115-40 in Sociologi-

cal Methodology 1990, edited by C.C. Clogg.Oxford: Basil Blackwood.

Comstock, George A. 1986. Public Communica-tion and Behavior, Vol. 1. New York: Aca-demic Press.

Cotton, John L. 1985. “Cognitive Dissonance inSelective Exposure.” Pp. 11-34 in SelectiveExposure to Communication, edited by D.Zillman and J. Bryant. Hillsdale, NJ: LawrenceErlbaum.

Efron, Bradley. 1987. “Better Confidence Inter-vals.” Journal of American Statistical Associa-tion 82:171-85.

Feldman, Jack J. 1966. The Dissemination ofHealth Information. Chicago: AldinelDe-Gruyter.

Flay, Brian R. 1987a. “Mass Media and SmokingCessation: A Critical Review.” American Jour-nal of Public Health 77:153-60.-. 1987b. Selling the Smokeless Society: 56

Evaluated Programs and Campaigns World-wide. Washington, DC: American Public HealthAssociation.-, Bonnie R. Brannon. C. Anderson

Johnson, William B. Hansen, Arthur L. Ulene,Deborah A. Whitney-Saltiel, Laura R. Gleason,Steve Sussman, Michael D. Gavin, Kim MarieGlowacz, Deborah F. Sobol, and Dana C.Spiegel. 1988. “The Television, School andFamily Smoking Prevention/Cessation Project I.Theoretical Basis and Television Program De-velopment.” Preventive Medicine 17:585-607.

__ and Dee Burton. 1990. “Effective MassCommunication Campaigns for Public Health.”Pp. 129-46 in Mass Communication for PublicHealth, edited by C. Atkin and L. Wallack.Newbury Park, CA: Sage.

Freedman, Jonathan L. and David 0. Sears. 1965.“Selective Exposure.” Pp. 57-96 in Advances inExperimental Social Psychology, Vol. 2, editedby L. Berkowitz. New York: Academic Press.

Gerbner, George, Larry Gross, Michael Morgan,and Nancy Signorelli. 1986. “Living withTelevision: The Dynamics of the CultivationProcess.” Pp. 17-40 in Perspectives on MediaEffects, edited by J. Bryant and D. Zillman.Hillsdale, NJ: Lawrence Erlbaum.

Green, Lawrence W., Marshall W. Kreuter, SigridG. Deeds, and Kay B. Partridge. 1980. HealthEducation Planning: A Diagnostic Approach.Palo Alto, CA: Mayfield.

Helman R., K.M. Cummings, B.P. Haughey,M.A. Zielezny. and R.M. O’Shea. 1991.“Predictors of Attempting and Succeeding atSmoking Cessation.” Health Education Re-search 6( 1):77-86.

Joreskog, Karl and Dag S&born. 1988a. USREL7: A Guide to the Program and Applications.Chicago: SPSS, Inc.

- a n d D a g S&born. 1 9 8 8 b . PRELJS.Mooresville, IN: Scientific Software.

Katz, Elihu. 1968. “On Reopening the Question ofSelective Exposure to Mass Communications.”Pp. 788-96 in Theories of Cognitive Consis-tency: A Sourcebook. edited by R. Abelson andE. Aronson. Chicago: Rand McNally.-. 1980. “On Conceptualizing Media Ef-

fects.” PD. 119-41 in Studies in Communica-tion, Vol. 1, edited by T. McCormack.Greenwich, CT: JAI Press.

Klapper, Joseph T. 1949. The Eficts of the MassMedia. New York: Columbia University Press,Bureau of Applied Social Research.-. 1960. The Effects of Mass Communica-

tion. Glencoe, IL: Free Press.Kviz, Frederick J., Kathleen S. Crittenden, Laurie J.

Belzer, and Richard B. Wamecke. 1991. “Psy-chosocial Factors and Enrollment in a TelevisionSmoking Cessation Program.” Health EducationQuarterly 18:445-61.

Liang, Jersey, Hitoshi Asano, Kenneth A. Bollen,Eva F. Kahana, and Daisaku Maeda. 1987.“Cross-Cultural Comparability of the Philadel-phia Geriatric Center Morale Scale: An Ameri-can-Japanese Comparison.” Journal of Geron-tology 42~37-43.

MacCallum, Robert. 1986. “Specification Searchesin Covariance Structures Modelling.” Psycholog-ical Bulletin 100: 107-20.

Manfredi, Clara and Richard B. Wamecke. 1984.“Level and Sources of Cancer Knowledge by .Demographic Characteristics and Cancer Site.”University of Illinois at Chicago. Unpublishedmanuscript.

-9 Loretta Lacey and Richard B. Warnecke.1990. “Results of an Intervention to ImproveCompliance with Referrals for Evaluation ofSuspected Malignancies at Neighborhood PublicHealth Centers.” American Journal of PublicHealth 80( 1):85-87.

McGuire, William J. 1968. “Selective Exposure:A Summing Up.” Pp. 797-800 in CognitiveConsistency: A Sourcebook, edited by R.Abelson and E. Aronson. Chicago: RandMcNally.-. 1985. “Attitudes and Attitude Change.”

Pp. 233-346 in Handbook of Social Psychology,Vol. 2. edited by G. Lindzey and E. Aronson.3rd edition. New York: Random House.

!

I

I

j

I

i

HEALTH BEHAVIOR CHANGES THROUGH TELEVISION 335

1986. “The Myth of Massive MediaGet: Savagings and Salvagings.” Pp. 175-

Research, Vol. 2, edited by T. Kinnear. AnnArbor, MI: Association for Consumer Research.

259 in Public Communication and Behavior, Snyder, Mark and William Ickes. 1985. “Person-Vol. 1, edited by G. Camstock. New York: ality and Social Behavior.” Pp. 883-948 in TheAcademic Press. Handbook of Social Psychology, Vol. 2, edited

Novotny, Thomas E., Kenneth E. Warner, Juliette by G. Lindzey and E. Aronson. 3rd edition.S. Kendrick, and Patrick L. Remington. 1989. New York: Random House.“Smoking by Blacks .and Whites: Socioeco- Sobel, Michael E. and George W. Bohmstedt.nomic and Demographic Differences.” Ameri- 1985. “Use of Null Models in Evaluating the Fitcan Journal of Public Health 78: 1187-89. of Covariance Structure Models.” Pp. 152-78 in

Real, Michael R., Hayes L. Anderson, and Sociological Methodology 1985, edited by S.

Marilyn H. Harrington. 1980. “Television Leinhardt. San Francisco: Jossey-Bass.

Access for Older Adults.” Journal of Communi- Sweeney, Paul D. and Kathy L. Gruber. 1984.

cation 3081-88.“Selective Exposure: Voter Information Prefer-

Roberts, Donald F. and Nathan Maccoby. 1985.ences and the Watergate Affair.” Journal of

“Effects of Mass Communication.” Pp. 539-96Persona& and Social Psychology 46: 1208-2 1.

in The Handbook of Social Psychology, Vol. 2,Wamecke, Richard B., Brian R. Flay, Frederick J.

edited by G. Lindzey and E. Aronson. Reading,Kviz, Charles L. Gruder, Patricia Langenberg,Kathleen S. Crittenden, Robin J. Mermelstein,

MA: Addison-Wesley.Sears, David 0. 1968. “The Paradox of De Facto

Marlene Aitken, Siu Chi Wong. and Thomas D.

Selective Exposure without Preferences forCook. 1991. “Characteristics of Participants in a

Supportive Information.” Pp. 777-87 in Theo-Televised Smoking Cessation Intervention.”Preventive Medicine 20:289-303.

n’es of Cognitive Consistency: A Sourcebook, Wamecke, Richard B., Patricia Lange&erg, Siuedited by R. Abelson and E. Aronson. Chicago: Chi Wong, Brian R. Flay, and Thomas D.Rand McNally. Cook. 1992. “24-Month Follow-up of Partici-

.-, and Jonathan L. Freedman. 1967. “Selec- pants in the Second Chicago Televised Smokinguve Exposure to Infon‘nation: A Critical Re- Cessation Program.” American Journal of Pub-view.” Public Opinion Quarterly 3 I: 194-213. lit Health 82(6):835-40.

Smith, Ruth B. and George P, Moschis. 1984. Zillman, Dolf and Jennings Bryant (eds.). 1985.“Consumer Socialization of the Elderly: An Selective Exposure to Communication. Hills-Exploratory Study.” In Advances in Consumer dale, NJ: Lawrence Erlbaum.

Brian R. Flay is director of the Prevention Research Center and professor in the School of Public Health,University of Illinois at Chicago. His research interests include smoking cessation, AIDS prevention, andviolence prevention.

Stephanie M&d1 is assistant professor, Department of Health Promotion Sciences, College of PublicHealth, University of Oklahoma Health Sciences Center, Oklahoma City. Her research interests includethe evaluation of community health promotion interventions.

Dee Burton is associate director for media research of the University of Illinois at Chicago PreventionResearch Center. Her work focuses on the role of the mass media in health.

Thomas D. Cook is professor of sociology, psychology, education and social policy, NorthwesternUniversity. His areas of interest are. the design and execution of social experiments, methods forpromoting causal generalization, and theories of evaluation practice.

Richard B. Wnrnecke is director of the Survey Research Laboratory and professor of sociology andepidemiology/biostatistics, University of Illinois at Chicago. He has been engaged in smoking research inlow SES populations for the last 15 years.