Men’s Preferences for Physical Activity Interventions: An Exploratory Study Using a Factorial Survey Design Created With R Software

Abstract

Effective exercise interventions are needed to improve quality of life and decrease the impact of chronic disease. Researchers suggest males have been underrepresented in exercise intervention studies, resulting in less understanding of their exercise practices. Findings from preference survey methods suggest reasonable association between preference and behavior. The purpose of the research described in this article was to use factorial survey, a preference method, to identify the characteristics of exercise interventions most likely to appeal to male participants, so preferences might be incorporated into future intervention research. The research was guided by the framework of Bandura’s social cognitive theory, such that variations in individual, environmental, and behavioral factors were incorporated into vignettes. Participants included 53 adult male nonadministrative staff and contract employees at a public university in the Southeastern United States, who each scored 8 vignettes resulting in 423 observations. Multilevel models were used to assess the influence of the factors. Participants scored vignettes that included exercising with a single partner, playing basketball, and exercising in the evening higher than vignettes with other options. Qualitative analysis of an open response item identified additional alternatives in group size, participant desire for coaching support, and interest in programs that incorporate a range of activity alternatives. Findings from this research were consistent with elements of social cognitive theory as applied to health promotion. Factorial surveys potentially provide a resource effective means of identifying participants’ preferences for use when planning interventions. The addition of a single qualitative item helped clarify and expand findings from statistical analysis.

Introduction

Regular participation in moderate to vigorous exercise is believed to increase life span and decrease risk for many types of chronic disease (U.S. Centers for Disease Control and Prevention [CDC], 2014). There are both advantages and challenges associated with an exercise-based approach to disease prevention. Advantages include potential to accrue mental and physical health benefits (CDC, 2015); challenges include that researchers have incomplete understanding of the causal factors that affect ongoing participation in exercise (Bauman et al., 2012). Intervention research presents an opportunity for researchers and practitioners to increase knowledge of the correlates of exercise participation (Freene, Waddington, Chesworth, Davey, & Cochrane, 2014; Purath, Keller, McPherson, & Ainsworth, 2013), although some researchers (e.g., George et al., 2012; Waters, Galichet, Owen, & Eakin, 2011) have reported that males are consistently underrepresented in exercise intervention research. While there are notable examples of successful evidence-based exercise interventions, including those with more male than female adherents (e.g., Cadmus-Bertram et al., 2014), or those developed to appeal to male participants (e.g., Hunt, McCann, Gray, Mutrie, & Wyke, 2013), George et al. (2012) and Waters et al. (2011) recommended that researchers and practitioners continue to improve their understanding of the factors that might motivate and encourage men’s initiation and ongoing participation in regular exercise, in order to ensure that more males have access to potential health-promoting benefits.

A number of researchers have solicited participants’ input on design of exercise interventions. Focus areas have included preference for group versus individual activities (e.g., Beauchamp, Carron, McCutcheon, & Harper, 2007), preferred delivery method for intervention information (e.g., Vandelanotte et al., 2013), or activity preferences (e.g., Nies, Troutman-Jordan, Branche, Moore-Harrison, & Hohensee, 2013). Of these, activity preference may be most important; according to Iso-Ahola and St. Clair (2000), individuals who adhere to exercise participation have tended to select activities that provided them with a “sense of control and competence” (p. 136). Parfitt and Gledhill (2004) concluded that choice of activity resulted in an immediate positive impact on mood and less perceived fatigue, but there is limited evidence from exercise intervention research about the impact of stated choice or preferences on acquisition of or adherence to exercise behavior. Researchers from other health professions have concluded there is a reasonable to strong relationship between a priori stated choice and behavior in some instances. Ryan and Watson (2009) compared two choice research methods and reported that participants’ stated choice for a Chlamydia screen under varied conditions, when compared with actual behavior, was a match 78% to 80% of the time. Lambooij et al. (2015) assessed parents’ preferences for infant hepatitis B screening given multiple varied circumstances, and found stated choice predicted actual positive behavior 85% of the time although the match between choice and negative behavior (i.e., choosing not to obtain the vaccine) was much less at 26%.

The purpose of the research described in this article is to assess potential utility of a preference research method, called factorial survey, to contribute to researcher and practitioner knowledge about what features males might find more or less appealing in an exercise intervention. The eventual aims following this exploratory study are to incorporate men’s expressed preferences into intervention design, and to evaluate preference-based designs to provide evidence for or against this approach for intervention planning.

This research was guided from a theoretical standpoint by Bandura’s (1986) social cognitive theory (SCT), from an applied practice standpoint by the principles of evidence-based practice (EBP), and from a resource-savings standpoint by use of a sophisticated survey research method to pilot test intervention characteristics. A further goal of this research was to develop an efficient method to administer and manage the survey process through use of open access and commonly available software programs.

Social Cognitive Theory

According to Bandura (1986), behavior reflects a complex interaction among individual factors, environmental factors, and the characteristics of the behavior itself. Bandura (2004) also described the importance of various self-regulatory practices in health behavior change. In keeping with SCT, the survey instrument was designed to assess environmental and behavioral components, to provide participants with opportunities to describe variations on these, and to comment on individual factors in an open response item. Additionally, use of a multilevel statistical model facilitated identification of variability at both the sample and individual participant level.

Evidence-Based Practice

Although the evidence component of EBP is often emphasized, according to Sackett et al. (1996), evidence-based medical practice represents a combination of “individual clinical expertise and the best available external evidence” (p. 71) such that both are necessary. Within the area of clinical expertise, Sackett et al. (1996) included a more pronounced role by patients or clients in making care or treatment decisions. This research was guided by a desire to respect and accommodate the values and preferences of potential intervention research participants, combined with a practical expectation, based on literature cited above from other disciplines, that participants’ expressed preferences are frequently associated with actual behavior. Other components of EBP, including review of available high-quality evidence, and reliance on researcher expertise, were incorporated through review of existing exercise intervention research, and through incorporation of items in the survey that reflected the researchers’ personal experiences, observations, and findings from preliminary research (Chatfield & Hallam, 2015).

Method

Factorial surveys provide a means to assess simultaneously multiple characteristics of a situation through use of a series of descriptive passages known as vignettes. Rossi and Anderson (1982) described the factorial survey process as a blending of experiments and surveys. According to Lauder (2002), because several circumstances are bundled together, factorial survey vignettes more closely approximate real life than traditional surveys. Although researchers have used factorial surveys to explore judgments, beliefs, and intentions (Wallander, 2009), the method has only been infrequently used for intervention or program planning purposes (e.g., Hennessy, MacQueen, & Seals, 1995), despite the fact that it reflects a potentially resource effective approach for gathering information (Lauder, 2002).

Vignettes in factorial surveys consist of either random or deliberate¹ combinations of various levels of multiple quantitative, categorical, or dichotomous predictor variables. Respondents’ scoring of several vignettes makes regression analysis possible, and the estimated coefficient values represent the relative weight of the predictor variables.

While authors have offered some considerations for vignette designs (e.g., Auspurg, Hinz, & Liebig, 2009; Shooter & Galloway, 2010), the technological process of creating an instrument is not often detailed. Some older research studies refer to obsolete software, while recent authors have described use of currently available software (e.g., Brauer et al., 2009) that requires substantial financial investment. There is a further limitation for some researchers in that; many programs, including SAS macros, were developed for use only on the PC platform.

In order to create a more accessible process, the open access multiplatform software program R was used in combination with Microsoft Office® software programs to efficiently create a large number of random vignettes (800) for administration through paper and pencil surveys. Appendix A includes an abbreviated description of the process and sample R code.

The researchers derived the list of varying conditions presented in the vignettes by considering application of SCT to exercise behavior, considering findings from previous interview research conducted with male exercisers (Chatfield & Hallam, 2015), reviewing typical programming described in exercise intervention literature, and identifying available or changeable circumstances. According to Bandura (1986), individuals require support to make positive change including acquisition of goal setting skills; this was incorporated through use of the Coaching dimension. Also, according to Bandura, environmental factors can be both influential and limiting; in this worksite study, availability of flexible working hours was identified by the research team as both a relevant and potentially changeable aspect of environment. The researchers selected dimensions referencing individual preference (Activity types, Group sizes, and Time) based on a combination of availability and review of prior exercise intervention research.

Consistent with practices from prior factorial survey, research respondents were directed to assess how James, a generic middle-aged male who was not currently a regular exerciser, would respond, rather than to express respondents’ personal preferences. According to Freidenberg, Mulvihill, and Caraballo (1993), use of a third person can counter any participant distrust of researchers or research. Hughes and Huby (2002) described use of a third person in vignettes additionally as advantageous for distancing participants from potentially uncomfortable subject matter. According to Caro et al. (2011), who cited support provided by factorial survey originators Rossi and Anderson, it can be assumed that participant’s recommendations for the vignette subject are reflective of the participants’ own preferences.

The scoring for this instrument was modeled on the factorial survey intervention research conducted by Hennessy et al. (1995), and worded as a probability item (i.e., How likely is it that James would begin to exercise if this program was offered?). Participants were provided with a 0 to 100 scale of identical length for each vignette, with marks in increments of 5 and labeled at the ends and center point. Participant responses that did not fall on one of the lines were measured with a ruler, and a score was assigned, rounded to ½ point. Each participant was given a survey packet that included eight unique randomly created vignettes and, following the work of Ganong and Coleman (2006), a single open response item. The directions for the open response item prompted respondents to provide their opinions about the ideal exercise program for James. A sample survey excerpt that includes a vignette is contained in Appendix B.

An institutional review board approved the research project, and the survey was pretested for clarity of directions and content with a subset of the sample of interest. The only change made as a result of the pretest was provision of additional clarifying directions for the qualitative item. The institutional review board approved the amended instrument.

Participants

This research project was planned as an exploratory study to develop community-based exercise interventions for the priority population that consisted of adult males, in particular those who were irregularly active or not active. A public university is one of the largest area employers in a small rural town in which the research took place, which suggests it has potential as a point of access to many residents. Based on concerns expressed by human resources personnel about exercise opportunities for staff, the researchers targeted male university staff who did not fall under faculty, executive, or paraprofessional categories, and university contract employees, who held similar classifications, as the convenience sample for this exploratory research project.

Factorial survey research is generally analyzed using a multilevel regression model. Some authors (e.g., Hox, 2010; Kreft & de Leeuw, 2007) have provided rule of thumb estimates and described more complex formulas for sample size and power calculations to be applied with multilevel regression, although as Auspurg and Hinz (2015) point out, ample simulation studies have not been completed to determine how well such calculations work when applied to factorial survey designs, that are substantially more complex than standard multilevel regression models. Hox (2010) and Kreft and de Leeuw (2007), both emphasized number of groups over number of members to maximize power; for the purposes of this research, participants, who each assessed multiple vignettes, comprised the group level. Therefore, the researchers adopted Hox’s rule of thumb recommendation to target 100 groups (participants) in order to best estimate random effects, variance, and covariance. Hox further recommended group size of 10. The researchers balanced this recommendation with recommendations from research conducted to identify optimal number of vignettes each participant could reasonably rate in a session (i.e., Sauer, Auspurg, Hinz, & Liebig, 2010) and reduced the number of vignettes to eight. This resulted in a target number of 800 observations.

In keeping with institutional policies, human resources personnel managed recruitment through unit managers, and recruitment was planned to continue just until the desired number of surveys was completed. Unit managers described the survey to staff in person, and consent wording was included in the survey packet. Participants were informed that their participation was both confidential and voluntary. Unit managers, who demonstrated varying levels of interest in the research process, had final control over distribution and completion of surveys.

Surveys were produced in pencil and paper format because many participants did not regularly use computers during their working hours. Demographic information requested included age in 10-year bands, self-identified race or ethnicity, and an estimate of how regularly respondents currently participated in exercise. Participants indicated age as follows: 15 or 28% indicated ages 18 to 29; 17 or 32% indicated ages 30 to 39; 12 or 23.6% indicated ages 40 to 29; 8 or 15.2% indicated ages 50 to 59; the remaining participant identified as ages 60 to 69. Seventeen or 32% of participants self-identified as Black; 34 or 64% or participants self-identified as White, 1 participant self-identified as Hispanic, and 1 self-identified as other.

The organization’s human resources department requested that participants provide a rough estimate of current activity/inactivity levels, so the survey was modified to include a general checklist. Eight or 15.5% of participants reported currently exercising 5 or more days per week. Eleven or 20.8% of participants reported exercising 3 to 4 days per week. Fourteen or 26.4% of participants reported exercising 1 to 2 days per week. Some participants reported no regular exercise during the past 1 to 2 months (5 or 9.4%), or no regular exercise during the past 6 months (14 or 26.4%); one participant reported no regular exercise during the past 3 to 4 months. These responses are summarized in Table 1.

Table 1.

Participant Demographic Information, n = 53.

Characteristic	n/%
Age (years)
18-29	15/28.3
30-39	17/32
40-49	12/23.6
50-59	8/15.2
60-69	1/1.9
Race/ethnicity
Black	17/32
Hispanic	1/1.9
White	34/64
Other	1/1.9
Reported current exercise
5 Days or more per week	8/15.1
3-4 Days/week	11/20.8
1-2 Days/week	14/26.4
No regular exercise; past 1-2 months	5/9.4
No regular exercise; past 3-4 months	1/1.9
No regular exercise; past 6 months	14/26.4

Results

Statistical Analysis

Respondents returned 57 of the distributed surveys although final analysis was based on 53 surveys comprising 423 observations; 4 respondents failed to verify age over 18 years and 1 respondent neglected to score the last program in his survey packet. Predictors included in the model were Activity (basketball, jogging, swimming, tennis), Group (alone, group of 25, group of 8-12, combination of alone and in a group, with one partner), Time of Day (evening, midday, morning), Work (ability to work flexible hours or not), Coaching (no coaching, coaching for goal setting once a week, once a month, when contacted), and Health (family history of heart disease, high blood pressure, no current health concerns, joint pain or arthritis, overweight). The first author entered scores and, expanding the recommendation of Guttmacher, Kelly, and Ruiz-Janecko (2010) that 10% of data entry be checked, a graduate student checked data entry on a randomly selected sample of 13% of the surveys; no corrections were indicated.

All statistical analyses were conducted using R (R Core Team, 2013) and the nlme package (Pinheiro, Bates, DebRoy, Sarkar, & R Core Team, 2013). The distributionof the outcome variable was approximately normal, slightly platykurtic, and nearly symmetrical (mean = 49.37; median = 50; kurtosis = 2.28; skew = 0.09). Hierarchical linear regression models were used for analysis due to violation of the assumption of independence that results when each participant accounts for multiple observations (Hox, Kreft, & Hermkens, 1991). The calculated intraclass correlation of 24%, a measure of the proportion of variance accounted for when data are analyzed as nested, further supported a multilevel approach. Following directions provided by Hox (2010) and Bliese (2013), the authors assessed the intercept and the contribution of each predictor as either fixed or random. All models other than the initial intercept-only comparison were run using restricted maximum likelihood estimation.

A chi-square difference test between the −2 log likelihood of the fixed and random intercepts models had a statistically significant result (χ² = 45.98; degrees of freedom = 1; p < .001), which indicated better model fit with random intercepts. Model fit was improved (χ² = 28; degrees of freedom = 14; p = .014) by allowing the slope for Group to vary randomly. According to Hox (2010), changes in variance estimates can serve as a proxy for R² to assess overall model fit (i.e., How much inclusion of predictors improves the ability of the model to predict when compared with a no predictor model) for multilevel regression models. Adding all predictors accounted for an additional 4% of the between-group (difference between respondents) variance, and 6.4% of the within-group (how each individual scored his eight vignettes) variance when compared with the intercept-only model. The variance change attributable to the addition of the random slope was calculated using an adjustment formula provided by Hox and attributed to Snijders and Bosker. This result suggested that 44.86% of the within-group variance was accounted for by including a random slope for the predictor variable Group, which included the options for how many exercise partners were preferred.

According to Kreft and de Leeuw (2007), hypothesis testing of individual parameters, in particular for social science researchers, is less helpful than a focus on overall fit of the proposed model. Consistent with this guidance, all parameters were retained in the final model. Table 2 reports dimensions, sources and parameter estimates with standard errors, confidence intervals, and probability values. Additional detail and R code used for this statistical analysis are available by contacting the first author.

Table 2.

Parameter Estimates and Confidence Intervals, n = 423 (L1), 53 (L2).

Parameter	Source	Level	Estimate (SE)	Lower 95% CI	Upper 95% CI	p
Intercept			55.78 (5.21)	45.53	66.03	<.001**
Activity (reference cell: basketball)	Available community activities	Bicycling	−1.60 (3.69)	−8.87	5.66	.67
		Jogging	−10.07 (3.49)	−16.94	−3.21	<.05*
		Swimming	−6.94 (3.45)	−13.73	−0.16	<.05*
		Tennis	−12.88 (3.82)	−20.38	−5.37	<.001**
Group (reference cell: alone)	George et al. (2012); Oka, King, and Young (1995)	Group of 25	5.5 (4.03)	−2.43	13.44	.173
		Group of 8-12	5.36 (3.77)	−2.06	12.77	.156
		With a group once per week; alone otherwise	−0.5 (3.77)	−7.92	6.92	.90
		With one partner	11.1 (4.21)	2.79	19.38	<.001**
Time (reference cell: in the evening)	Prior interview research	In the middle of the day	−2.0 (2.41)	−6.73	2.74	.41
		In the morning	−5.23 (2.35)	−9.86	−0.61	<.05*
Work (reference cell: blank)	Prior interview research	Can work flexible hours to fit in exercise	1.57 (1.95)	−2.27	5.41	.42
Coach (reference cell: blank)	Conn et al. (2011); Annesi (2002); Annesi and Unruh (2007)	Coach helps set goals once a week	1.97 (2.75)	−3.44	7.37	.47
		Coach helps set goals once a month	−0.6 (2.99)	−6.48	5.29	.84
		Coach helps when contacted	−0.12 (2.69)	−5.41	5.17	.96
Health (reference cell: family history of heart disease)	Researcher interest	High blood pressure	−0.86 (3.23)	−7.21	5.49	.80
		No current health concerns	1.67 (2.93)	−4.1	7.44	.57
		Joint pain or arthritis	−6.06 (3.11)	−12.18	0.05	.05
		Overweight	−5.4 (3.11)	−11.52	0.73	.08

Note. SE = standard error; CI = confidence interval.

^*p < .05. **p < .001.

Statistically significant predictors included options within the predictors Activity, Group, and Time of Day. This suggested that these options had most influence on how individuals scored the scenarios. Most preferred options were basketball (in Activity), with one partner (in Group) and evening (in Time of Day). Alternatives within Coach (whether a health coach was available), Work (whether time off was given during the work day for exercise), and Health (whether there was a preexisting health concern), were not statistically significant, which suggested these details exerted less influence over how participants scored the vignettes. The authors also created a visual representation that depicts approximations of the relative ratings of the options within each category, including those that were and were not statistically significant, based on parameter estimates; see Figure 1.

Figure 1.

Visual approximations of participant ratings.

Qualitative Findings

Thirty participants responded to the open response item. Qualitative coding was conducted employing open coding methods that included descriptive and in vivo coding (Saldaña, 2013), facilitated by use of the comment/track changes functions of the Microsoft Office software program Word. Codes were applied to “meaningful qualitative units” (Chenail, 2012, p. 268), which consist of the smallest excerpt from data that can be seen to express meaning. These codes were organized and collapsed into six categories, matching like with like, based on the authors’ assessment of the qualitative similarity of the content. The first author derived the categories and then all authors considered code assignments. Discrepancies in code assignment were resolved through discussion among authors to reach consensus; the original sources in their entirety were relied on to provide clarifying information in instances of disagreement. After category assignment was finalized, the authors rereviewed both the categories and representative data excerpts, and collectively collapsed the categories into three broader themes. Excerpts contained in each theme were then reassessed one additional time in context of the respondent’s entire response, to further assess fit under the assigned theme. Theme labels were developed in an effort to represent and describe the data in a meaningful and engaging way. Additional detail regarding the qualitative analysis, including deidentified data excerpts, is available by contacting the first author.

Developed themes included the following: easing into it, choice and changes, pressure versus persuasion. The theme easing into it included examples provided by several respondents in which they recommended that James be encouraged to approach his increase in exercise incrementally. According to one respondent, it is necessary to “take it easy with someone then work your way up,” while another described the need for James to increase “speed and exercise time steadily as he goes.” An additional respondent expressed concern that any of the programs shown in his sample of vignettes might be overwhelming for James as a new exerciser, and he might instead need to begin with slow walking.

The survey contained only one activity alternative that could be described as cross-training (weight training 2 days per week and jogging 3 days per week) but several respondents used the open response section to describe plans that incorporated two or more activities per week. These recommendations comprised the theme choice and changes. Respondents provided plans that included alternative types of exercise (e.g., “stretching and cardio,” “weights and cardio on alternating days”) or suggested alternating similar types of exercises (e.g., “jogging and bicycling,” “bike, swim, and play basketball”). Respondents in some instances provided health-related reasons for activity alternatives (e.g., “light weights to help with weight loss”), although one of the repeated reasons given for variation in activities was “so he [James] won’t get bored.”

Coaching was mentioned frequently in the open responses. The theme incorporating these excerpts was titled pressure versus persuasion because the responses reflected opposing approaches. Some respondents felt that it was the place of the coach to “push” or to be in “constant contact” with James so someone would be “holding [him] accountable.” One participant mentioned that James would be most likely to succeed in a program that was “similar to what is done in the military.” Another respondent cautioned that the role of the coach is to “help him set goals” but specifically directed “don’t push.” Another respondent described this in similar terms: “[to] help him set goals and help him achieve goals.” Other respondents seemed to combine the role of a workout partner with some of the functions of a coach, observing that the best workout partners can “motivate you,” and that “involving other people makes you exercise.”

Discussion

These research findings provide initial guidance for exercise interventions and suggest that size of group, type of activity, and provision of support are items of concern. These findings also demonstrate that inclusion of a single open response item both added to and clarified the results of statistical analysis. While statistical analysis suggested that some activity alternatives were less well received than others, open responses from participants helped clarify this. For instance, the parameter estimate for swimming was substantially lower than the reference cell basketball, although multiple respondents specified swimming as an alternative in the open response item, subject to certain conditions. One respondent accompanied a high rating for a program including swimming with the written condition: “if he [James] has help.”

Parallels to the qualitatively developed themes can be found in published research reports. The idea of easing into exercise was explored by Hunt et al. (2013) who reported that male participants in their research responded positively to a walking program when it had the added appeal of being presented in a sports-focused context (i.e., the program introduction and initial training was held in league soccer stadiums). Research on cross-training, or alternating among several activity choices, has tended to focus on the utility of variety in encouraging performance improvement (e.g., Issurin, 2013) rather than on encouraging initiation or adherence to exercise, although several authors of performance research additionally acknowledged the contribution of cross-training to combat boredom or to improve overall fitness (Foster et al., 1995; Ruby et al., 1996).

The role of others in exercise has been assessed through examination of the negative association between exercise participation and self-described loneliness (Hawkley, Thisted, & Cacioppo, 2009), through assessment of the positive association between larger social networks and likelihood to meet exercise requirements (Marquez et al., 2014), and through a systematic review that reinforced the utility and encouraged further use of peer mentors to facilitate exercise participation (Petosa & Smith, 2014). Additionally, Griffith, King, and Allen (2012) reported positive influence of peers on men’s exercise participation was frequently identified in qualitative focus group research conducted with Black participants. The finding of Griffith et al. (2012) is of particular interest as George et al. (2011) identified a lack of published exercise intervention research with male participants who represent racial or ethnic minorities. Although multiple researchers have suggested that companionship and encouragement plays some role in exercise adherence, as a result of earlier qualitative research findings (Chatfield, 2015), the first author has speculated that role might be of greater importance for new exercisers.

Respondents also commented about size of workout group, and generally advocated very small groups (e.g., two, three, or five other people), which were not included as options of the survey. Group sizes represented on the survey instrument included a single partner, followed by a group of 8 to 12, selected by the authors to represent a small group, in contrast to the third option which was a large group of 25 or more. That respondents had preference for a group size larger than a single partner but smaller than the small group option provided suggests that this item might be better represented in future surveys by including more alternatives.

The Health Coach parameter did not achieve statistical significance, and based on this alone could be presumed less influential. The qualitative findings addressing this item suggest, instead, that coaching is subjectively defined concept that merits additional exploration. Several participants identified both coaching and workout partners as sources of support, which implies these items might be viewed as interchangeable, although participants differed regarding whether they thought coaches should motivate through encouragement or through discipline.

Findings from this exploratory study offer some support for Bandura’s (2004) applications of SCT toward health promotion to frame this and future work. In general, respondents who provided qualitative data expressed that James would be successful at transforming himself into a regular exerciser, given the appropriate environmental facilitators. Multiple participants alluded to the role of goal setting, described as a critical part of the self-regulatory process by Bandura. According to Bandura (2004), individuals’ health behavior improvements are encouraged through setting “short term attainable goals” (p. 145). One of the participants in this research expressed this as the importance of setting “manageable goals.” In the longer term, Bandura (2004) noted “motivation is enhanced by helping people to see how habit changes are in their self interest and in the broader goals they value highly” (p. 144). Another research participant noted the importance for James’s adherence that he experience recognition for broader goals “such as weight loss or lowered blood pressure.”

Given that the proportion of U.S. adult males accruing regular participation in moderate to vigorous exercise, based on movement sensor data, is relatively low and decreases greatly with age (Troiano et al., 2008), while risk for chronic disease increases, this focus on men’s preferences to inform intervention design seems to be warranted. While the findings from this exploratory research provided less definitive information for intervention design than was ideal, there are three specific areas that, as a result of inconsistent qualitative and quantitative findings, merit further study to better understand and represent the alternatives of greatest interest. These are the role of mentoring or coaching, identification of optimal group size, and the appeal of offering a range of different activities to comprise proposed interventions. What these findings do suggest that might be immediately useful for researchers and practitioners is that these respondents preferred to exercise with a partner, or perhaps a small group, considered evening a preferred time of the day, and basketball, as well as cross-training alternatives, were well received activity choices. The activity alternatives likely were influenced by what respondents viewed as available alternatives in their environment.

Limitations of this research include that respondents’ expressed preference is not necessarily a proxy for choice or adherence to a program, although it might reasonably be assumed that respondent’s choices were more likely to reflect preferences and less likely to reflect disinterest or dislike. Given the rating mechanism, it was possible to assign a 0 score to all programs and no participant did so. Additional limitations include smaller than desired sample size along with difficulty of using power analysis methods in a factorial survey design to identify an optimal sample size. Purposive and convenience sampling greatly limits generalizability as does the fact that the sample included employees of a single organization, who represented limited employment categories. It is also possible that use of a D optimal design (see Note 1) would have resulted in superior parameter estimates. Finally, alternatives contained in the vignette were driven more by the researchers’ knowledge regarding available resources than by support from prior research, or the results of a preliminary needs and interest assessment.

Despite these limitations, the findings considered in total suggest that factorial survey provides a potentially viable and efficient alternative to gather information from participants to inform designs of intervention research. Given the results of this research study, further assessment and reporting of factorial survey administration and analysis, and future evaluation of interventions developed as a result of factorial survey research is warranted.

Appendix A

Producing Random Factorial Survey Vignettes With R and Microsoft Office

1. Install the R base package, create working directory, and install the add on xlsx (Dragulescu, 2015)

2. Develop predictor variables and identify levels (see a simplified version of this survey research in the following table)

3. Create and run R code (a sample follows). This code will create 400 vignettes; 8 are allocated to each of 50 participants.

4. Retrieve and open the Excel spreadsheet from working directory; delete first column (sequential numbering) from the spreadsheet.

5. Create the stem items and scoring bar in Microsoft Word; use Word form letter/merge function to place column headers from the spreadsheet in the form letter. Also include a space for participant and program numbers; this facilitates matching up the scores with the lines on the spreadsheet.

6. Print and sort surveys by participant number, attach other documents (consent, demographic items), and distribute surveys.

7. Create a score column in the Excel spreadsheet and enter participant scores, matching up participant/program numbers. Save spreadsheet as comma separated values (.csv) file.

8. Read the .csv file into R. R will read the category levels as categorical variables; the default is to set the reference cell based on alphabetical order of category names.

9. Run statistical models and diagnostic tests as necessary

Activity	Group	Frequency	Time	Coaching
Walking and jogging	Alone	Three times per week	In the morning	None
Bicycling	With one partner	Four times per week	At midday	Weekly
Walking and weight training on alternate days	In a group of 8-12	Five times per week	In the evening	Monthly
Playing tennis	In a group of 25		Varying times of the day	As desired

#SAMPLE R CODE TO PRODUCE VIGNETTES

#installing the add on to write the data to Excel

install.packages("xlsx")

library(xlsx)

#1 creating the groups

activity<-c(“walking and jogging”, “bicycling”, “walking and weight training on alternate days”, “playing tennis”)

group<-c(“alone”, “with one exercise partner”, “in a group of 8 to 12”, “in a group of 25”)

frequency<-c(“3 times per week”, “4 times per week”, “5 times per week”)

time<-c(“in the morning”, “at midday”, “in the evening”, “at varying times of the day”)

coach<-c(“The health coach will help James set weekly goals”, “The health coach will help James set monthly goals”, “The health coach will help James set goals whenever James wants to contact the coach“,“ ”)

#2 drawing the samples from the first group

Participant<-rep(1:50, each = 8)

Program<-rep(1:8, 50)

set.seed(1984)

fsActivity<-replicate(400, sample(activity, length(1), replace = T))

fsGroup<-replicate(400, sample(group, length(1), replace = T))

fsFrequency<-replicate(400, sample(frequency, length(1), replace = T))

fsTime<-replicate(400, sample(time, length(1), replace = T))

fsCoach<-replicate(400, sample(coach, length(1), replace = T))

#3 putting the random items together in a data frame and create an #Excel workbook in the working directory

fsFrame<-data.frame(Participant, Program, fsActivity, fsGroup, fsFrequency, fsTime, fsCoach)

write.xlsx(fsFrame, “factorialSurvey.xls”, col.names = T)

Appendix B

Sample Vignette

James is 45 and does not exercise on a regular basis. He would like to start.

On the following pages, there are eight different versions of an exercise program. In some of these scenarios, James also has specific health concerns.

I would like to know how likely you think James would be to begin to exercise regularly if he was offered each of these programs, based on your knowledge and experience. Some of these programs include equipment (e.g., bicycles) or facilities (e.g., pools). You can assume for the purposes of this survey that James will be given equipment or facility use if he chooses those programs.

For each program, you may assign a score between 0 and 100, with 0 meaning that you think James is not likely to try this program and 100 meaning that you think James would be very likely to begin to exercise regularly if he was offered this program. You may go back and revise your ratings at any time. Indicate your score by making a mark anywhere on the line. If you want to change your score, you may erase or X out your prior score, or circle your revised score.

After you have considered all of these programs, there is a space at the end of this form for you to describe the exercise program you think would be most likely to inspire James to exercise regularly. If you complete this portion, you can use any of the features from the programs contained in this survey, or you can suggest your own ideas. You do not need to complete this section if you do not wish to. After you have finished, please place your completed pages in the envelope provided.

Survey # X

Program Y

The activity is bicycling.

• James will be exercising with one partner.

• James will be exercising in the morning.

• James can work flexible hours to fit in his exercise.

• The health coach helps James set goals once a week.

• James has a family history of heart disease.

Between 0 and 100, with zero meaning not at all likely and 100 meaning very likely, how likely is James to beginning exercising if Program Y is offered?

(Place a mark anywhere on the line)

Not

At all                                             Very

Likely                                            likely

0____|____|____|____|____|____|____|____|____|____50____|____|____|____|____|____|____|____|____100