Abstract: Hypotheses generated by the precaution adoption process model, a stage model of health behavior, were tested in the context of home radon testing. The specific idea tested was that the barriers impeding progress toward protective action change from stage to stage. An intervention describing a high risk of radon problems in study area homes was designed to encourage homeowners in the model’s undecided stage to decide to test, and a low-effort, how-to-test intervention was designed to encourage homeowners in the decided-to-act stage to order test kits. Interventions were delivered in a factorial design that created conditions matched or mismatched to the recipient’s stage (N = 1,897). Both movement to a stage closer to testing and purchase of radon test kits were assessed. As predicted, the risk treatment was relatively more effective in getting undecided people to decide to test than in getting decided-to-act people to order a test. Also supporting predictions, the low-effort intervention proved relatively more effective in getting decided-to-act people to order tests than in getting undecided people to decide to test.
Most current theories of individual health behavior consist of a set of variables thought to be important and a rule (or equation) prescribing how these variables should be combined (Conner & Norman, 1996; Weinstein, 1993). However, a number of researchers have questioned whether reactions to health hazards can be represented adequately by a single prediction rule. Instead, they describe the adoption of. precautions in terms of a series of stages (Baranowski, 1992-1993; Horn, 1976 ; Janis & Mann, 1977; Prochaska & DiClemente, 1983; Weinstein, 1988; Weinstein & Sandman, 1992).
Article Table of Contents
The most distinctive and potentially useful feature of stage theories is the idea that the determinants of progress toward protective action vary from stage to stage. The factors most important in getting someone to first pay attention to a risk, for example, may not be the ones that are most important in determining whether he or she eventually decides to take action. Thus, stage theories imply that treatments need to be matched to the stage of the audience, focusing on the specific barriers that inhibit movement to the next stage and changing over time as the audience progresses from stage to stage (DiClemente, Carbonari, & Velasquez, 1992). By suggesting how to tailor interventions to audiences, stage theories offer the prospect of more effective and more efficient behavior change efforts.
Most non-stage theories, in contrast, are based on a single theoretically or empirically derived equation (e.g., Ajzen & Madden, 1986; Fishbein & Ajzen, 1975; Ronis, 1992). This equation generates a numerical value for each person, and this value is interpreted as the likelihood that the person will take action. The prediction equation thus places each person along a continuum, and the goal of interventions is to move people along the continuum. Such an approach acknowledges quantitative differences among people, but it does not admit the possibility of qualitative changes in the barriers that interfere with progress. The notion of matching interventions to people is either incidental or completely missing in this approach.
The Precaution Adoption Process Model (PAPM)
This present article describes an experimental test of a particular stage theory, the precaution adoption process model (Weinstein, 1988; Weinstein & Sandman, 1992). The PAPM distinguishes among seven stages:
- unaware of the health action;
- aware but not personally engaged;
- engaged and trying to decide what to do;
- decided not to act (a step out of the sequence toward action);
- decided to act but not yet having acted;
- acting; and
- maintaining the new health-protective behavior (Weinstein & Sandman, 1992).
Thus, the proposed sequence of stages leading to action is 1-2-3-5-6-7. The model asserts that people usually pass through this sequence in order, without skipping any stage, although there is no minimum length of time that must be spent in any one stage. Movement backward toward an earlier stage can also occur.
Some data have been published pertaining to the PAPM (e.g., Blalock et al., 1996; Weinstein & Sandman, 1992) and a much larger amount pertaining to another stage theory, the transtheoretical model (for an overview, see Prochaska et al., 1994). The great majority of these data come from cross-sectional surveys and compare the attributes of people in different stages. Although substantial between-stages differences are often found, differences among people in different stages can be present even if behavior change is actually a continuous process (Weinstein, Rothman, & Sutton, 1998). Cross-sectional data provide only weak support for stage models.
A few experiments based on stage ideas have been conducted (Campbell et al., 1994; Prochaska, DiClemente, Velicer, & Rossi, 1993; Skinner, Strecher, & Hospers, 1994). However, they tend to compare a highly personalized, stage-based intervention with a standardized intervention, so it is not clear whether it is the content of the stage-based intervention or simply its personalization that makes it more successful.
Still, rigorous experimental tests of stage theories are possible (Weinstein et al., 1998). Experiments can directly test the prediction that Treatment I is better than Treatment II for moving people from stage A to stage B, whereas Treatment II is better than Treatment I for moving people from stage B to stage C. The 2 (preintervention stage) × 2 (interventions) design suggested by this example is the simplest experimental test of the fundamental idea. that different issues are important at different stages. Crossing two transitions with two properly selected interventions creates two conditions in which the interventions and stages are matched and two in which they are mismatched. This is the core of the design adopted for our experiment, although we add a control condition (no treatment) and a combination intervention (Treatment I + Treatment II), leading to a 2 (preintervention stage) × 2 (presence or absence of Treatment I) × 2 (presence or absence of Treatment II) design.
Examining the PAPM in the Context of Home Radon Testing
The PAPM is a framework that claims to identify important stages along the route to action. Experimental support for a stage framework requires one to find treatments that have different effects at different stages. It should be kept in mind, however, that transitions at each stage may be influenced by several variables (Weinstein, 1988) and that the variables that are important will depend, to some extent, on the nature of the health behavior.
The health behavior examined in our investigation was home radon testing. Radon is a radioactive gas produced by the decay of small amounts of naturally occurring uranium in soil. Radon in homes is the second leading cause of lung cancer after smoking (National Academy of Sciences, 1988; U.S. Environmental Protection Agency [EPA], 1992a), causing an estimated 7,000 to 30,000 deaths a year in the United States (Puskin, 1992). Radon concentrations exceeding the EPA’s recommended action level are present in about 6% of homes in the United States (EPA, 1991, 1992b). At present, reducing radon in existing homes is left largely to voluntary action. Individuals must decide on their own that testing is desirable and, if excessive radon is found, that radon reduction is needed.
Testing for radon does not involve ingrained habits or arduous and repetitive procedures. A positive radon test does not mean one is afflicted with a fatal illness. For these reasons, a brief intervention might be enough to change people’s decisions and actions.
In this experiment we focused on two transitions relevant to radon testing: from being undecided about testing one’s home (stage 3) to deciding to test (stage 5), and from deciding to test (stage 5) to actually ordering a test (stage 6). Two interventions were needed, one matched to each transition. We did not study the transition from never having thought about testing to thinking about testing because merely participating in a radon study and answering questions about testing would probably be sufficient to produce this change. We excluded people who had decided not to test because a brief intervention would probably be unable to reverse that decision.
Previous surveys and experiments (Sandman & Weinstein, 1993; Weinstein, Sandman, & Roberts, 1990) suggested that increasing homeowners’ perceptions of personal risk – that is, increasing the perceived likelihood of having unhealthy home radon levels in their own homes – is an important factor in getting undecided people to decide to test. This was chosen as the focus of one intervention.
Interventions focusing on risk have not been effective, however, in getting people to order tests (Weinstein et al., 1990; Weinstein, Sandman, & Roberts, 1991). Instead, several studies have found that increasing the ease of testing increases the number of test orders (Doyle, McClelland, & Schulte, 1991; Weinstein et al., 1990, 1991). Thus, for people who had already decided to test, we developed an intervention that would lower the barriers to action by providing information about do-it-yourself test kits and a test order form.
The basic hypothesis examined here was that interventions matched to stage would be more effective than interventions that were mismatched. In other words, the crucial test was for an interaction between stage and intervention. Additional hypotheses accompany the data analysis. We did not investigate how many stages exist or whether people pass through the stages in the specified sequence. Comparisons of stage predictions with the predictions of other theories of health behavior will be reported elsewhere.
Overview of Study Design
Pilot interviews and focus groups revealed that people in the study area knew relatively little about radon. Their beliefs appeared to reflect vague recollections of the issue as it had appeared in the media years earlier. Because their stated testing stage and testing intentions might be weakly held and unstable, all participants viewed a general informational video before receiving any experimental treatment. Their stage of testing was assessed by questionnaire after this first video (preintervention measurement). Only people in the undecided or decided-to-act stages with respect to radon testing took part in the succeeding experiment
Within a week after the questionnaire had been returned and eligibility to continue had been determined, the experimental interventions were delivered to participants. One intervention (High Likelihood) focused on increasing the perceived likelihood of having a home radon problem. The second (Low Effort) focused on decreasing the perceived and actual effort required to test. These two treatments were combined factorially to create four conditions: Control (no intervention), High Likelihood, Low Effort, and Combination (High Likelihood + Low Effort). Study participants were assigned at random to one of these four experimental conditions, creating a 2 (preintervention stage: undecided or decided-to-test) × 2 (High Likelihood treatment: present or absent) × 2 (Low Effort treatment: present or absent) factorial design. Beliefs about radon were assessed by questionnaire after the experimental treatment (postintervention measurement), and a follow-up interview was used to determine final stage and actual test orders.
A site was sought that had both elevated levels of radon (so that messages claiming high risk would be justified) and limited previous attention to radon (so that residents would be receptive to new information on this issue). Columbus, Ohio, met these criteria. Radon levels there were high (64–75% of homes were above EPA’s suggested action level [EPA, 1993; Grafton, 1990; H. E. Grafton, personal communication, April 1995]), and no major attempts to encourage radon testing had occurred since a campaign 8 years earlier.
People with listed telephone numbers in the Columbus, Ohio, area were screened on several variables. Telephone interviewers asked to speak to “the man or woman in your household who is most likely to make decisions about home environmental hazards, such as asbestos or lead-based paint.” Screening criteria were as follows: (a) Owns a single-family house or townhouse; (b) has heard of radon; (c) has not tested for radon; (d) has never thought about testing, is undecided, or plans to test; (e) has no plans to move in the next year; (f) owns a videocassette recorder; and (g) is 65 years of age or less. (Pilot studies showed that older people were less likely to test, claiming, with some validity, that reducing radon after many years of exposure would probably have little effect.)
Three different videos were developed for the experiment, using student ratings, focus groups of Columbus residents, and finally a large-scale pilot study in Columbus to refine the messages and their presentation.
Initial educational message
All participants viewed a 6-min tape titled “Basic Facts About Radon,” which provided a brief overview of the topic. It explained what radon is, how it causes illness, how it gets into houses, and how radon problems can be solved. The narrator stated that elevated radon levels have been found throughout the United States, but no details were provided about the risk in any particular location. The video noted that “the Surgeon General and the American Medical Association join the EPA in strongly recommending home radon testing.” It further stated that “kits that homeowners can use themselves are also available and easy to use. A wide range of test kits are on the market.” However, no information was provided about the advantages of specific types of tests, their cost, how to use the test kits, or how to obtain them.
High Likelihood condition
This treatment consisted of a 5-min video, “Radon Risk in Columbus Area Homes,” and an accompanying cover letter. The goal of the video was to convince people that they have a moderate to high chance of finding unhealthy radon levels in their own homes. Results of radon studies indicating high local levels, pictures of actual local homes with high levels, and testimony by a local homeowner and a city health official all presented evidence of the problem. The geological explanation for the high regional levels was illustrated with maps, and myths about radon levels that had been identified in past research were presented and refuted. Radon testing was briefly mentioned in one 11-s segment: “Fortunately, it is much easier than you may suppose to find out whether your home has a radon problem. Low-cost test kits are readily available and simple to use.”
Low Effort condition
Participants in this condition received a 5-min video, “How to Test Your Home for Radon,” an accompanying cover letter, and a form to order test kits through the American Lung Association (ALA). The video described how to select a kit type (making an explicit recommendation in order to reduce uncertainty), locate and purchase a kit, and conduct a test. The process was represented as simple and inexpensive. The order form was developed in collaboration with the National Safety Council and the ALA of Mid-Ohio and bore their emblems. It offered the short-term test kits recommended in the video at $7 each, and long-term test kits at $16 each. These were the same prices that the ALA and the National Safety Council charged the general public. This video said nothing about the frequency or seriousness of high radon levels.
People in this condition received a 10-min video that was simply a combination of the “Radon Risk in Columbus Area Homes” and “How to Test Your Home for Radon” video segments, in that order. They received the same letter and order form as people in the Low Effort condition.
Participants in this group received a letter stating that their assistance in viewing a second video was not needed (they had already screened “Basic Facts About Radon”).
Letters accompanying each experimental condition video (but not the “Basic Facts” video) provided limited information about obtaining home test kits. All letters stated, “More information about radon can be obtained from the American Lung Association of Mid-Ohio, which is listed in the white pages. The American Lung Association office also sells inexpensive home radon kits.” In the Low Effort and Combination conditions, the letters added, “For your convenience, a form for ordering a kit is enclosed.” The letter received by participants in the Control condition contained the same statement about the availability of information and tests from the ALA as the cover letter accompanying the High Likelihood video.
Video response questionnaires
Accompanying all videos was a questionnaire made up of three parts. Part A focused on the organization of the video content. It also asked participants, “On what two topics would you be most interested in having more information?” There were two choices relating to each of four issues: the health effects of radon; the likelihood of finding high levels in one’s home; radon testing; and radon reduction. Part B focused on the visual and sound quality of the video. Part B and the video organization questions of Part A were included to be consistent with our request for respondents’ reactions to the videos and are not discussed further.
Part C asked respondents for their thoughts about radon after having viewed the video. Among the questions in Part C was a sequence designed to assess stage of testing: “What are your thoughts about testing your home for radon?”: (a) “I have already completed a test, have a test in progress, or have purchased a test” [tested stage]; (b) “I have never thought about testing my home” [not-engaged stage]; (c) “I’m undecided about testing my home” [undecided stage]; (d) “I’ve decided I don’t want to test” [decided-not-to-test stage]; or (e) “I’ve decided I do want to test” [decided-to-test stage]. People who selected the last response were then asked , “Is testing something you plan to do soon or just something you’ll do someday when you get a chance?” Intentions to test were assessed with the question, “How likely would you say it is that you will test your home for radon in the next few months?” The response options were 0 (definitely won’t), 1 ( probably won’t), 3 (50-50 chance), 4 ( probably will), and 5 ( definitely will).
One set of additional questions asked participants about risk issues: the likelihood of finding radon problems in their own homes (1 = very unlikely, 5 = very likely); the same issue in percentage terms (1 = less than 10%, 5 = greater than 90%); and the percentage of homes in the Columbus area with radon problems (1 = less than 10%, 5 = greater than 90%). A second set of questions referred to the ease of testing: “How easy do you think it would be for you to locate and buy a do-it-yourself kit...?” (1 = very difficult, 4 =very easy); “How easy do you think it would be for you to use a radon test kit? (1 =very difficult, 4 = very easy); and “What do you think is the typical cost of a home radon test kit?” (1 = under $10, 5 = over $100). Participants were also asked about the difficulty of reducing home radon levels (1 = very difficult, 4 = very easy).
This brief telephone interview assessed participants’ final stage, whether they had ordered and conducted a radon test in their home, and what radon topics they wanted more information about. To detect possible false reports of testing, those who said they had tested were asked what type of kit they had purchased, where they obtained the kit, how long it remained in place, and so on.
Recruitment interviews were completed with 24,484 Columbus area residents. A total of 41.5% screened out because they were not homeowners, not in the desired age range, because of other factors unrelated to radon, or a combination of these. Of the remaining, 3.8% had never heard of radon, 20.6% had never thought about testing, 20.4% were undecided, 18.1% had decided not to test; 7.9% had decided to test, and 29.2% had already tested. When those eligible at this point in the study (those who had never thought about testing, were undecided, or had decided to test) were invited to react to the videos about radon that we had developed, 65.0% agreed.
Those consenting (N = 4,706) were mailed the video, “Basic Facts About Radon,” and a questionnaire assessing their reactions. Those individuals who were either in the undecided stage or decided-to-test stage after watching “Basic Facts About Radon” were assigned at random to one of the four experimental conditions and were mailed the materials appropriate for that condition. To enhance the impact of the interventions, participants were asked to watch each video twice: once for content and once for style. The response rate to the second video was 73.2%, with no significant differences among conditions.
Follow-up telephone interviews (completion rate = 94.5%) were carried out 9–10 weeks after respondents returned the second video questionnaire, allowing them ample time to have ordered a test. All those who completed the follow-up interview were mailed a thank you letter and a radon test kit order form.
Participants were dropped from the data analysis if they were missing important data (n = 18), said that someone else in their household had watched the experimental video (n = 15), said they had no role in “making decisions about things like radon testing” (n = 1), or could not answer questions about the test they claimed to have purchased (n = 2). This left a sample of 1,897 (58.8% women and 41.1% men). Preintervention (i.e., after “Basic Facts About Radon”), the division among stages of those retained in the study was 28.8% undecided and 71.2% decided-to-test.
The median length of residence was 7 years, and 52.5% of the sample was between 36 and 50 years of age. College graduates comprised 77.1% of the group. The racial distribution was 91.3% White and 5.4% African American, with smaller portions in other groups. A comparison of the sample with the population from which it was drawn was not possible because demographic data restricted to owners of single-family homes were not available.
Although age, sex, education, years of residence in their present home, presence of children under age 10 in the home, and smoking status were all assessed, only education proved to be appreciably correlated with subsequent test orders. More education was associated with greater testing (r = .10, p < .0001). None of the interactions between demographic variables and stage, experimental condition, or intentions affected testing, so demographic variables are not discussed further, although analyses were conducted while controlling for education.
Manipulation Checks and Preintervention Differences Between Stages
As would be expected from random assignment, there were no significant differences among conditions on any preintervention variable.
Perceived radon risk
Preintervention and postintervention means were examined to see whether the High Likelihood intervention had produced the effect intended. The three risk questions (perceived likelihood in own home, percentage chance in own home, and percentage prevalence in community) were combined to form a Perceived Risk scale ( = 83; range = 1–5). The High Likelihood intervention produced highly significant and nearly identical increases in perceived risk in the High Likelihood and Combination conditions: from 2.28 to 3.39, t (471) = 29.3, p < .0001, and from 2.33 to 3.49, t(468) = 30.5, p < .0001, respectively. There was a slight decline in perceived risk in the Low Effort condition: from 2.19 to 2.09, t(445) = 3.8, p < .002.
Supporting the idea that deciding to test is partly determined by perceptions of personal risk, mean perceived risk before the intervention was 1.97 and 2.40 in the undecided and decided-to-test groups, respectively, F(1, 1885) = 130.4, p < .0001, with stage accounting for 6.5% of the variance in the risk variable. In contrast, stage accounted for only 0.0%–0.8% of the variance in the questions about finding test kits, using test kits, fixing radon problems, and the cost of testing.
The effects of the interventions on perceived risk were independent of respondents’ preintervention stage, and there were no Preintervention Stage × Treatment interactions (ps > .4).
Perceived ease of testing
As intended, the Low Effort intervention succeeded in convincing people that testing is quite easy (combined ease of finding a test kit and ease of using a test kit: = .62; range = 1–4). The increases in the Low Effort. and Combination conditions were nearly identical: from 3.34 to 3.77, t(445) = 17.9, p < .0001, and from 3.27 to 3.75, t(468) = 16.0, p < .0001, respectively. There was a very slight decrease in the High Likelihood condition: from 3.40 to 3.35, t(471) = 2.4, p < .02. The perceived cost of testing, another indicator of ease, declined in the Low Effort and Combination conditions: from $26.46 to $17.40, t(398) = 10.3, p < .0001, and from $25.87 to $17.85, t(424) = 9.3, p < .0001, respectively, with negligible change in the High Likelihood condition (from $25.85 to $26.41, ns). (Dollar values were calculated using the midpoint of the range for each scale choice, with over $100 interpreted as $125.)
The effects of the interventions on perceived ease of testing were independent of respondents’ preintervention stage, and there were no Preintervention Stage × Treatment interactions (ps > .4).
Reported desire for information
Both pre- and postintervention, participants indicated the topics about which they would like more information. Variables were created to indicate whether they had expressed an interest in the likelihood of radon problems in local homes (risk), testing (testing), health effects (health), and home radon reduction (reduction).
There were no preintervention differences between conditions in the proportion of the sample interested in any topic. As expected, after treatment there were highly significant between-conditions differences on all four information topics, F(2, 1414) = 133, 238, 26, and 54, for risk, testing, health, and reduction information, respectively (all ps < .0001), but there were no Differences by Preintervention Stages or Preintervention Stage X Condition interactions (ps > .1). Collapsing across preintervention stage, the High Likelihood intervention was found to decrease desire for risk information (from 79% to 16% and from 76% to 18% in the High Likelihood and Combination conditions, respectively). In a similar vein, interest in testing information decreased in the Low Effort and Combination conditions (from 33% to 3% and from 29% to 5%, respectively). Thus, as intended, the High Likelihood treatment appeared to satisfy respondents’ needs for information about personal risk, and the Low Effort treatment appeared to satisfy respondents’ needs for information about how to test.
Although participants in both stages showed a high level of interest in information about the local prevalence of radon problems after viewing the “Basic Facts” video, this interest was higher among undecided (80.8%) than decided-to-test (74.2%) participants, F(1, 1895) 94, p < .003. Undecided participants also showed greater interest in information about health effects, 45.6% vs. 36.1%, F(1, 1895) = 14.8, p <.0001. In contrast, decided-to-test participants showed greater interest in information about testing: 33.0% versus 21.6%, F(1, 1895) = 24.5, p < .0001; and about radon reduction, 23.3% versus 18.9%, F(1, 1895) = 4.5 p < .05. Of course, showing that interest in information varies with stage is not the same as showing that receiving this information will lead to a change in stage.
Predicting Progress Toward Action
From a stage perspective, interventions are successful if they move people to any stage closer to action, so action is not the only or even the most appropriate measure of the effectiveness of an intervention. First, we examine progress toward testing as the dependent variable, then we examine test orders themselves.
Table 1 shows the percentage of people from each preintervention stage who progressed one or more stages toward testing. We chose this outcome criterion for several reasons. First, even though the lack of some specific information may prevent people from moving to the next stage, they might already possess the information needed to go from the next stage to the one following. Consequently, when people stuck at a particular stage are helped by an intervention to advance to the next stage, some of these people may proceed additional stages without further assistance. Second, all participants in our study received some information that could lower barriers to carrying out a decision to test. The “Basic Facts” video briefly informed them about the existence of easy-to-use, do-it-yourself tests, and the cover letter sent to participants in all conditions told them where to obtain inexpensive test kits. Thus, the High Likelihood treatment might help people decide to test, and the information given about testing, although limited, might be sufficient to allow some of these people to proceed to the next stage and order tests.
|Decided to test|
|Note: n = number of people in each experimental group.|
The upper half of Table 1 indicates the percentage of people at follow-up who had moved from the undecided stage to either the decided-to-test or the testing stage. The lower half of the table shows the percentage of decided-to-test people who had moved on to the testing stage. Although the table has each of the four conditions in a separate column, the analysis of variance used a 2 × 2 × 2 statistical model that included pretreatment stage, High Likelihood treatment, Low Effort treatment, and their interactions. (Because of the large sample sizes, the results of statistical analyses using log-linear, procedures are completely indistinguishable from the results presented here, despite the fact that progress is a dichotomous variable. Results derived from analysis of variance procedures are presented because of their greater familiarity.) (Note 1)
The analysis showed more people progressing from the undecided stage than from the decided-to-test stage, F(1, 1886) = 6l.6, p < .0001, and more progress from those who received the High Likelihood treatment than from those who did not, F(1, 1886) = 31.5, p < .0001. Most important, as predicted, there was a significant Stage × High Likelihood Treatment interaction, F(1, 1886) = 18.5, p < .0001, indicating that the High Likelihood treatment was much more effective for undecided participants than for decided-to-act participants.
There was also a large main effect of the Low Effort treatment, F(1, 1886) = 89.4, p <.0001. The Stage × Low Effort Treatment interaction, F(1, 1886) = 5.9, p < .02, indicated that, as hypothesized, the Low Effort treatment in the Low Effort and Combination conditions had a relatively bigger effect on the people already planning to test than on people who were undecided. The High Likelihood × Low Effort interaction and the three-way interaction were not significant
Predicting Test Orders
Follow-up interviews indicated that radon tests were ordered by 342 study participants or 18.0% of the sample. Of this total, 63.4% were ordered from the ALA of Mid-Ohio; the rest were obtained from various stores, government offices, and private testing companies.
The experimental data concerning test orders are presented in Table 2. For people initially planning to test, progress means actually testing according to the PAPM, so the data in the lower half of Table 2 are the same as those in the lower half of Table 1. As expected, there was more testing from the decided-to-test stage than from the undecided stage, F(1, 1887) = 42.3, p < .0001. In addition, there was much more testing from people exposed to a Low Effort treatment than from those who did not receive this treatment, F(1, 1887) = 87.9, p < .0001. The High Likelihood treatment effect and the Low Effort × High Likelihood interaction were not significant (ps > .1). Most important was the highly significant interaction between Stage and Low Effort treatment, F(1, 1887) = 18.2, p < .0001. The other interactions (Stage × High Likelihood and Stage × Low Effort × High Likelihood) were not significant (ps > .1).
|Undecided||(a) 5.1||(b) 3.5||(c) 10.1||(d) 18.7|
|Decided to Test||(e) 8.0||(f) 10.4||(g) 32.5||(h) 35.8|
|Note: Values in table are percentages.|
Next, a series of more detailed tests examined the cell-by-cell contrasts that are predicted by the PAPM. Each of these predictions about test orders can be viewed as a planned comparison. In subsequent paragraphs, the predictions are presented in brackets, with experimental groups labeled by letters that refer to the cells in Table 2.
Test order rates of both undecided and decided-to-test participants in the Control condition were expected to be quite low because both groups were viewed as lacking information needed to progress to action [(a) APPROX= (e), both small]. The main problems facing people who had decided to test were hypothesized to be the difficulties in choosing, purchasing, and using radon test kits. Thus, the Low Effort treatment was expected to be much more helpful than the High Likelihood treatment in getting people in this stage to actually order tests [(g) > (f)]. In fact, past research (Weinstein et al., 1990, 1991) suggested that the High Likelihood treatment would be ineffective in eliciting testing from people planning to test [(f) APPROX= (e)], and, more obviously, unable to elicit test orders from undecided people [(b) APPROX= (a)]. Furthermore, because it was anticipated that people in the decided-to-test stage did not need further information about risk, we predicted that testing in the combination condition would not be significantly greater than testing in the Low Effort condition [(h) APPROX= (g)].
According to the PAPM, people who are undecided have to decide to test before acting, so a Low Effort intervention alone was not expected to produce test orders from this group [(c) APPROX= (a)]. However, undecided people in the Combination condition received both High Likelihood information (seen as important in deciding to test) and Low Effort assistance (seen as important for carrying out action intentions). Some of these people might be able to make two stage transitions [(d) > (c)], but not as many as decided-to-test people in the Combination condition who only needed to advance only one stage [(d) < (h)].
We conducted t tests to compare the means of the cells mentioned in the preceding eight hypotheses. These demonstrated that none of the pairs predicted to be approximately the same were significantly different (ps > .3), but all pairs predicted to be different were significantly different: all ps > .0001, except for the hypothesis that (d) > (c), p > .03. We return to this last hypothesis in the next section.
Calculations of Two-Stage Transitions
The cell-by-cell hypotheses just presented had been based on the expectation that our interventions were completely stage specific. In particular, our initial expectation about the rate of testing from undecided people in the Low Effort condition (cell c) was based on the expectation that this treatment would not persuade anyone that they should test. Yet Table 1 shows that the Low Effort treatment did get undecided people to decide to test. Given this new knowledge, how should we revise our predictions? According to the precaution adoption process model, in order to test, undecided people have to make two separate stage transitions. If these transitions are viewed as independent sequential steps, the probability of a person moving forward two steps (from undecided to testing) should be the product of the two separate probabilities involved: the probability that undecided people move toward testing (and thus get at least to the decided-to-test stage) times the probability that people who get to the decided-to-test stage carry out this decision. According to this reasoning, the predicted rate of testing by undecided people in the Low Effort condition should be .364 × .325 = .118 or 11.8% (see Table 1 for the separate probabilities). This is very close to the observed value of 10.1% in Table 2. This same argument can be used to calculate testing rates for undecided people in the Combination condition. The expected rate of testing in this cell is .545 × .358 = .195 or 19.5%. This is, again, extremely close to the observed value of 18.7% in Table 2.
This research was stimulated by the belief that the adoption of new health-protective behaviors is usually too complex to be explained by a single prediction equation, no matter how many variables that equation contains. We suggested, instead, that there are relatively distinct stages along the path to action and offered the precaution adoption process model to summarize these stage notions. For health promotion, the most important contribution of a stage perspective is the idea that the obstacles inhibiting movement vary from one stage to the next. In this investigation we focused on two stage transitions, from undecided to decided to act and from decided to act to acting.
Perhaps the first point that should be emphasized in this discussion is the magnitude of the effects produced by our interventions. When viewed in terms of odds ratios – for example, the three-fold difference in test orders between the undecided and decided-to-test stages in the Low Effort condition or the 10-fold difference between cells with the highest and lowest testing rates – the effects observed here were quite large.
Second, one limitation of this study should be recognized. Although the PAPM is a theory of individual action, radon testing usually occurs in a family setting. The relationship between preintervention stage and testing behavior would probably have been even stronger if our participants had made their decisions without any other adult input. Our sample had too few participants living alone to test this idea.
Predictions of the Precaution Adoption Process Model
Not only were the predicted Stage × Intervention interactions associated with progress toward testing supported but so too were all eight detailed hypotheses about test orders. Risk vulnerability was, as expected, more important in getting people to decide to act than in getting them to carry out this decision. In fact, risk information had no apparent value in producing action among people who had already decided to test. The Low Effort intervention, in contrast, greatly aided people who had decided to act but was relatively less important among people who were undecided.
No prediction equation from any current theory of health behavior can produce this pattern of results. The theory of planned behavior (Ajzen, 1985, 1991; Ajzen & Madden, 1986), which proposes a positive interaction between intentions and control, can explain an increase in the effectiveness of the Low Effort intervention among people closer to action, but it cannot explain why increasing perceived risk led people to decide to test (i.e., increased testing intentions) but did not increase testing.
The interaction observed between stage and the High Likelihood intervention might have been even stronger if the intervention had been more effective in raising risk perceptions. Despite our attempts to counter the myths people use to deny their risk, there was evidence that participants tended to resist the risk message. People in the High Likelihood and Combination conditions were told that 73% of the homes in their community had high radon levels, but they estimated their own chances at only 54%. Optimistic bias – the tendency to rate one’s own risk as lower than peers’ risk – actually increased after exposure to the risk information. (Although people in these conditions increased their estimates of their own risk, they increased their estimates of their neighbors’ risks still more.) In the conditions receiving risk information, the own risk minus others’ risk difference score changed from –4.7% preintervention to –8.7% postintervention, demonstrating an increase in optimistic bias, t(911) = 5.01, p < .0001. The postintervention optimistic bias was greater among undecided than among decided-to-test participants, –10.52 versus –5.74, F(1, 1371) = 18.1, p < .0001.
The interaction between stage and the Low Effort intervention might also have been stronger if that intervention had been less effective in persuading undecided people to test. We can imagine study participants who were reluctant to test nevertheless telling themselves, “If it is really that simple and inexpensive, I might as well do it.” Many other health behaviors are not nearly so easy, and in other situations similar low-effort interventions may be insufficient to convince undecided people that they should act.
The unexpected success of the Low Effort intervention among undecided people led us to use a calculation predicated on the notion of a sequential stage model – the idea that the probability cf moving forward two stages is the product of the probabilities of moving through each transition separately – to calculate the testing rates of undecided people who received the Low Effort treatment. The agreement between the calculated and observed rates (11.8% and 10.1%, respectively, in the Low Effort condition and 19.5% and 18.7%, respectively, in the Combination condition) was quite impressive, providing further support for the presence of a stage process.
The Combination Condition
Because the combination treatment was effective among both undecided and decided-to-test participants, one might be tempted to conclude that the PAPM has not provided any new treatment ideas. “Just use the combination treatment,” someone might say. There are several flaws in this reasoning.
First, the combination treatment was approximately twice as long as each of its two components. Media time is expensive; speakers usually have a fixed length of time for their presentations; and audiences have a limited attention span. Thus, attempting to replace the Low Effort or High Likelihood interventions with their combination would involve substantial costs.
Second, people are likely to be more engaged by the treatment that matches their stage. For example, when not participating in a research study, people who are undecided about taking a precaution may not pay attention to the detailed procedural information they might need later to carry out that precaution. The data presented earlier concerning interest in information provide some support for this claim.
Third, among people who had decided to act, the superiority of the combination intervention was actually negligible. The Low Effort treatment produced just as many test orders; the risk information was superfluous. As we suggested in the introduction, and as our data suggest, stage models offer the prospect of more effective and efficient interventions than a scattershot, one-size-fits-all approach. Nevertheless, if only a single message can be given to a mixed-stage audience, the combination intervention would probably be the most appropriate.
Predictions derived from stage models are undoubtedly more accurate in some situations than in others. For example, when people are asked for their decisions concerning new hazards or new precautions (topics to which they have given little thought), their responses may say little about their eventual actions. In fact, we may have been fortunate in this investigation that a brief video designed to remind participants about a topic that had received little local attention in the past 8 years elicited reactions that could predict their subsequent behavior. Stage models of health behavior (and probably most other models of health behavior) are likely to be most accurate among people who have been exposed to the health issue recently in their daily lives.
Have we proved that people pass through distinct stages as they come to adopt precautions? No. Neither have we demonstrated that applying treatments in the sequence suggested by the PAPM is better than applying them in another sequence, the gold standard test for a stage theory (Weinstein et al., 1998). It is still possible that precaution adoption can be explained by a continuous equation, although this equation would have to be quite complex if some variables increase in importance and some decrease in importance depending on the values of still other variables.
Even if behavior is eventually explainable by a complex prediction equation, stages can still be extremely useful. They can help us by identifying regions in the multidimensional space of independent variables where specific variables are particularly influential. By categorizing people who had not yet acted into distinct subgroups, the precaution adoption process model helped us to identify important barriers to action. Furthermore, it, suggested how to match interventions to individuals and successfully predicted the effects of such matching. Although the evidence for stages may be limited at this time, the idea certainly deserves further attention.
Note 1. Because no data were gathered from the control group in the time between the “Basic Facts” questionnaire and the follow-up interview, the three other groups had one additional opportunity to drop out of the study. Analyses showed that dropout rates were weakly related to education (r = –.05, p < .02) and to the perceived cost of radon testing (r = .07, p < .01) but not to any other measured variables (including stage and intentions to test). Weighting the control group data by education and perceived cost in an attempt to match the attrition in the other experimental groups had completely negligible effects on statistical analyses. Consequently, only the unweighted results are reported.
This study was supported with funding from the National Cancer Institute (CA60890), the Environmental Protection Agency (X824392-01-0), and the New Jersey Agricultural Experiment Station (Project 26101).
We wish to thank Margo Ogé, David Rowson, and Dennis Wagner of the Environmental Protection Agency; Leyla McCurdy of the American Lung Association; Jane Ann Page and Cheryl Cooper of the American Lung Association of Mid-Ohio; Michael Pompili and Harry Grafton of the Columbus Ohio Health Department; Nyki Brandon Palermo of the National Safety Council; Fred Rehbein of Reel Resources; David West of Insul-Tech Radon; Mark Schulman and Adrienne Viviano of Schulman, Ronca, & Bucuvalas; and Sharon Johnson for their cooperation and assistance. We also thank Paul Lehrer, Ann O’Leary, Barbara McCrady, Mark Conner, Meg Gerrard, Alex Rothman, Stephen Sutton, and anonymous reviewers for suggestions during the planning of the research and the preparation of this article.
Ajzen, I. (1985). From intentions to actions: A theory of planned behavior. In J. Kuhl & J. Beckmann (Eds.), Action control: From cognition to behavior (pp. 11–40). Berlin, Federal Republic of Germany: Springer-Verlag.
Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, 50, 179–211.
Ajzen, I., & Madden, T.J. (1986). Prediction of goal-directed behavior: Attitudes, intentions, and perceived behavioral control. Journal of Experimental Social Psychology, 22, 453–474.
Baranowski, T. (1992–1993). Beliefs as motivational influences in stages in behavior change. International Quarterly of Community Health Education, 13, 3–29.
Blalock, S.J., DeVellis, R.F., Giorgino, K.B., DeVellis, B.M., Gold, D., Dooloy, M.A., Anderson, J.B., & Smith, S.L. (1996). Osteoporosis prevention in premenopausal women: Using a stage model approach to examine the predictors of behavior. Health Psychology, 15, 84–93.
Campbell, M.K., DeVellis, B.M., Strecher, V.J., Ammerman, A.S., DeVellis, R.F., & Sandler; R.S. (1994). Improving dietary behavior The effectiveness of tailored messages in primary care settings. American Journal of Public Health 84, 783–787.
Conner, M., & Norman, P. (1996). Predicting health behavior. Philadelphia Open University Press.
DiClemente, C.C., Carbonari, J.P., & Velasquez, M.M. (1992). Alcohol treatment mismatching from a process of change perspective. In R.R. Watson (Ed.), Drug and alcohol abuse reviews (Vol. 3, pp. 115–142). Totowa, NJ: Humana Press.
Doyle, J.K., McClelland, G.H., & Schulze, W. D. (1991). Protective responses to household risk: A case study of radon mitigation. Risk Analysis, 11, 121–134.
Fishbein., M., & Ajzen, I. (1975). Belief, attitude, intention and behavior: An introduction to theory and research. Reading, MA: Addison-Wesley.
Grafton, H.E. (1990). Indoor radon levels in Columbus and Franklin County, Ohio residences, commercial buildings, and schools. In The 1990 International Symposium on Radon and Radon Reduction Technology (EPA No. 600/9-90/00a, v. 1-Preprints, A-I-1).
Horn, D. (1976). A model for the study of personal choice health change. International Journal of Health Education, 19, 88–97.
Janis, I.L, & Mann, L. (1977). Decision making: A psychological analysis of conflict, choice, and commitment. New York: Free Press.
National Academy of Sciences (1988). Health effects of radon and other internally deposited alpha-emitters: BEIR IV. Washington, DC: National Academy Press.
Prochaska, J.O., & DiClemente, C.C. (1983). Stages and processes of self-change in smoking: Toward an integrative model of change. Journal of Consulting and Clinical Psycholology. 51, 390–395.
Prochaska, J.O., DiClemente, C.C.,Velicer, W.F., & Rossi, L.S. (1993). Standardized, individualized, interactive, and personalized self-help programs for stages of smoking cessation. Health Psychology, 12, 399–405.
Prochaska, J.O., Velicer, W.K., Rossi, J.S., Goldstein, M.G., Marcus, B.H., Rakowski, W., Fiore, C., Harlow, L.L., Redding, C.A., Rosenbloom, D., & Rossi, S.R. (1994). Stages of change and decisional balance for 12 problem behaviors. Health Psychology, 13, 39–46.
Pushkin, J.S., (1992). An analysis of the uncertainties in estimates of radon-induced lung cancer. Risk Analysis, 12, 277–285.
Ronis, D.L (1992). Conditional health threats: Health beliefs, decisions, and behaviors among adults. Health Psychology, 11, 127–134.
Sandman, P.M., & Weinstein, N.D. (1993). Predictors of home radon testing and implications for testing promotion programs. Health Education Quarterly. 20, 1–17.
Skinner, C.S., Strecher, V.J., & Hospers, H. (1994). Physicians’ recommendations for mammography: Do tailored. messages make a difference? American Journal of Public Health, 84, 43–49.
U.S. Environmental Protection Agency, Radon Division, Office of Radiation and Indoor Air (1993, Sept). EPA’s map of radon zones, Ohio. Washington, DC: Author.
U.S. Environmental Protection Agency, Office of Radiation Progams (1991). National residential radon survey: Statistical analysis, national and regional estimates: Volume I. Washington, DC: Author.
U.S. Environmental Protection Agency, Office of Radiation Programs, & U.S. Department of Health and Human Services, Centers for Disease Control (1992a). A citizen’s guide to radon (2nd ed.). Washington, DC: Author.
U.S. Environmental Protection Agency, Office of Radiation Programs (1992b, May). Technical support document for the citizen’s guide to radon (EPA 400-R-92-011). Washington, DC: Author.
Weinstein, N.D. (1988). The precaution adoption process. Health Psychology, 7, 355–386.
Weinstein N.D. (1993). Testing four competing theories of health-protective behavior. Health Psychology, 12, 324– 333.
Weinstein, N.D., Rothman, A.M., & Sutton, S.R. (1998). Stage theories of health behavior: Conceptual and methodological issues. Health Psychology, 17, 290–299.
Weinstein, N.D., & Sandman, P.M (1992). A model of the precaution adoption process: Evidence from home radon testing. Health Psychology, 11, 170–180.
Weinstein,. N.D., Sandman,. P.M., & Roberts, N.E. (1990). Determinants of self-protective behavior: Home radon testing. Journal of Applied Social Psychology, 20, 783–801.
Weinstein, N.D., Sandman, P.M., & Roberts, N.E. (1991). Perceived susceptibility and self-protective behavior: A field experiment to encourage home radon testing. Health Psychology, 10, 25–33.
Copyright © 1998 by Health Psychology