Outdoor Home Gardener Preferences for Environmental Attributes in Gardening Supplies and Use of Ecofriendly Gardening Practices (2024)

Survey and data collection.

We conducted a survey of home outdoor gardeners using the online survey platform offered by Qualtrics (Provo, UT). The list frame, provided by Qualtrics, was composed of randomly selected Tennessee residents aged 18 years or older who self-identified as outdoor gardeners. In June 2018, we conducted an online pretest survey in Qualtrics of 108 randomly selected respondents provided from a panel by Qualtrics who were Tennessee residents, aged 18 years or older, and self-identified as home gardeners. We modified the survey based on the pretest results, and fielded a full version of the survey in July 2018 (n = 612 responses, with 601 usable in our statistical analysis). The survey instrument was approved by the University of Tennessee Institutional Review Board (UTK IRB-18-04526-XM).

The survey asked respondents questions about their participation in outdoor home gardening, the importance of environmentally friendly attributes on purchase of gardening supplies, and participation in ecofriendly outdoor gardening practices. The survey also elicited information about the use of gardening information, attitudes about gardening and the environment, household expenditures, and gardener demographics.

Respondents were asked if they planted pollinator plants (Pollinators), composted (Composting), recycled gardening packaging (Recycling), used rainwater collectors (Rainwater), used organic gardening practices (Organic Gard), or had tested their garden soil (Soil Testing). A dummy variable identified positive responses (1 = yes, 0 = no). Opinions about the importance of ecofriendly attributes of gardening products was elicited with a series of Likert scale questions (1 = not important at all, 5 = extremely important). For the purposes of modeling, importance the gardener places on each of the six attributes within the MIMIC modeling framework (discussed later in this article), a dummy variable was created for each attribute such that if the respondent indicated they believed the attribute was somewhat or extremely important by selecting 4 or 5 on the Likert scale, the attribute was assigned “1,” and “0” otherwise. These attributes are displayed in Table 1 as Decr Need Fertil, Decr Need Pestic, Decr Need Water, Native Species, Prod Organically, and Recyclable Pkg.

Respondents were asked about sources of information they obtained for making home gardening decisions. These included social media/Internet; other gardeners; television, magazines, and newspapers; or extension/Master Gardener sources. Dummy variables were created for each Social Media/Internet, Other Gardeners, TV Magazines, and Extension/Master Gardener, in which the dummy variable was “1” if they used the particular source and “0” otherwise (Table 1).

The survey also included a section to collect information on demographics attributes. Demographic variables include respondent age (Age GT50), gender (Female), level of education (College Grad), place of residence (Rural), and annual expenditures on gardening supplies (Ann Expend).

Respondents were also asked a series of questions regarding attitudes toward gardening and the environment using a Likert agreement scale (1 = strongly disagree, 5 = strongly agree) (Table 1). These include self-perceived knowledge assessment for making environmental decisions (LackEnvKnow), perceived responsibility to future generations for the environment (RespFutGen), perceived home gardener impact on the environment (HomeGardEnvir), and preferences for growing their own food (GrowFood). A copy of the survey instrument is available from the authors on request.

MIMIC model.

We use a MIMIC model with correlated errors to estimate the relationship between respondent demographic characteristics, the propensity to adopt the gardening practices considered, and preferences for the garden supplies attributes examined here. MIMIC models consist of structural and measurement equations (Skrondal and Rabe-Hesketh, 2004). MIMIC models are useful when there are multiple outcomes reflecting underlying latent variables (or propensities), as well as multiple causes (i.e., demographic variables). The type of model used in this study, the MIMIC model, is similar to a multinomial logit model. The advantage of using a MIMIC model rather than a series of multinomial logits is that it admits arbitrary correlation between groups of questions that are related through latent variables. The relationships in the MIMIC model in this study are illustrated in Fig. 1.

The relationships between the 12 indicator variables and the two latent variables are then estimated with 12 logit regression equations (measurement equations). If the estimated coefficient on the latent variable in a practice’s or attribute’s measurement logit equation is significant, then the practice or attribute is considered a good indicator of the underlying latent variable.

The disturbance terms of the structural regressions, ${\varsigma }^P$ and ${\varsigma }^A$, and errors from their respective measurement logit regressions, ${\epsilon }_1^P,\dots ,{\epsilon }_6^P$ and ${\epsilon }_1^A,\dots ,{\epsilon }_6^A$ are assumed to be independent and identically distributed random variables with an expected value of zero and a constant variance θ_j = π²/3 (Dell’Anno and Schneider, 2003; Joreskog and Goldberger, 1975; Lambert et al., 2015). The parameter ρ is the correlation coefficient between the error terms ${\varsigma }^P$ and ${\varsigma }^A$.

The statistical software STATA (release 14; StataCorp, College Station, TX) was used to estimate the bivariate MIMIC model. Specifically, STATA’s generalized structural equation modeling procedure (gsem) was used to estimate the MIMIC model. The procedure uses maximum likelihood to estimate the model parameters. Estimation details can be found in the gsem documentation.

Using the estimated coefficients from the logit models in Table 2, along with Eqs. [3] and [4] and Eqs. [5] and [6], and the respondent data, we developed probability response curves for adopting each practice and preferring each product attribute across varying levels of the latent variables Ecopract and Envattrib, respectively. The probability curves illustrate which practices are more or less likely to be used and which attributes hold broader (or narrower) appeal among gardeners of varying propensities to use ecofriendly gardening practices and to prefer environmental attributes.

The same procedure (substituting Eq. [2] into Eqs. [5] and [6]) is used to calculate the reduced form for the probabilities of preferring each attribute as a function of the causal variables.

These reduced form probabilities were used to illustrate how different market profiles for home gardeners could affect these probabilities. Two gardener profiles were developed for each structural equation (two for Ecopract and two for Envattrib). The profile for a gardener who is very unlikely (very likely) to adopt the ecofriendly practices is Profile 1 (Profile 2). The profile for a gardener who is very unlikely (very likely) to prefer the environmental attributes is Profile 3 (Profile 4). The profiles are developed based on the signs of the estimated coefficients (${\text{β}}^P$ and ${\text{β}}^A$) for the causal variables (X) in the estimated structural regression equations in Table 2. Because the signs of the coefficients on latent variables in the measurement logits are positive, this means the sign of the effect of the causal variable on the probability of using the practices or preferring the attributes will be determined by the sign on the coefficient of that causal variable in the structural regressions for Ecopract and Envattrib. For example, the estimated coefficient on the variable for female gender (Female) is negative in the structural equation for Ecopract; therefore, female gender has a negative effect on probabilities of gardeners using the ecofriendly practices. Hence, to project probabilities of using the ecofriendly practices using Eqs. [9] and [10], for Profile 1 gardeners, a value of “1” is used for Female, whereas for a Profile 2 gardener, a value of “0” is used for Female. A similar process is used in setting the values for each of the causal variables in each profile. The demographic characteristics, expenditures, information use, and attitudes for each profile are presented. Distributions around the predicted probabilities were simulated with 15,000 draws each using the Krinsky-Robb (Krinsky and Robb, 1986). The 50th percentile and 95% confidence intervals were calculated for the simulated probabilities of choosing the practices by Profiles 1–2 and choosing the attributes as important by Profiles 3–4. If the 95% confidence intervals for Profiles 1 and 2 do not overlap, then the probability of Profile 2 using the practices is statically greater than Profile 1. The same type of comparison can be made for the probability of choosing the attributes as important by Profiles 3 and 4.

Respondent summary statistics.

Approximately 37% of the respondents were older than 50 (Table 1). Most (79%) respondents were women. Approximately 40% of the respondents were college graduates. From the National Gardening Association 2014 survey, the average U.S. gardener was female, 45 years or older, and held a college degree or had some college education (National Gardening Association, 2014). The present sample appears to correspond with these National Gardening Survey averages. In 2018, the average amount spent per U.S. household on lawn and gardening activities was $503, whereas the present study found an average expenditure on gardening supplies of $297 for 2018 (Cohen, 2018). It should be noted that our estimate is for Tennessee outdoor home gardeners, not a U.S. average. Also, our question asked specifically about gardening supplies and did not include lawn care services or supplies expenditures, hence the survey respondent’s estimates of their expenditures would likely be lower than the U.S. estimate that does include lawn care.

The gardening practice used most frequently was Recycling (53%), followed by Pollinators (41%) (Table 1). The least commonly used practices were rainwater collectors, Rainwater (23%) and Soil Testing (9%). Among the attributes evaluated in this study, gardening products that decreased the amount of pesticides applied (Decr Need Pestic) was cited most frequently as important (89%), followed by decreased need for fertilizer, Decr Need Fertil (75%). The least commonly cited attributes were Native Species (59%) and Prod Organically (59%). For information sources used by home gardeners, the most commonly used source was Social Media/Internet (56%), followed by Other Gardeners (50%). Although 38% used TV Magazines, 31% used Extension/Master Gardener sources.

Home gardener opinions about the environment and gardening were elicited with a Likert agreement scale, where 1 = strongly disagree, and 5 = strongly agree. Concerning environmental knowledge, respondents tended to disagree that they lacked sufficient knowledge to make environmental decisions (LackEnvKnow, mean = 2.8). Respondents also tended to have very strong views that we have a responsibility to steward the environment for future generations (RespFutGen, mean = 4.3). Respondents also tended to agree that home gardeners can affect the environment (HomeGardEnvir, mean = 3.4), and that they like to garden to grow their own food (GrowFood, mean = 3.9).

Estimated MIMIC model.

The log-likelihood ratio test (LLR) of the model against the intercept-only model indicates that the covariates and latent variable were significantly related with the propensity to use certain garden practices and preferences for the garden supplies attributes analyzed here (Table 2). The estimated logit measurement equations are shown in the upper portion of Table 2. Each of the estimated coefficients on the latent variables Ecopract and Envattrib (${\text{λ}}^p$ and ${\text{λ}}^a$) are significant. This result suggests that the six practices are good indicators of the propensity to use ecofriendly gardening practices, Ecopract, and that the six attributes are good indicators of the propensity to prefer ecofriendly attributes, Envattrib, in gardening supplies. The percentage of observations correctly classified for each of practices indicator ranges from 71% for Pollinators to 91% for Soil Testing. The percentage of observations correctly classified by the MIMIC logits for each of the attribute indicator variables ranges from 73% for Prod Organically to 92% for Decr Need Pestic.

Several of the estimated coefficients on demographic and other causal variables (defined in Table 1) are significant in the structural equation for Ecopract (Table 2). Contrary to findings by Behe et al. (2010) and Grebitus et al. (2017), but similar to those by Campbell et al. (2020), female gender has a negative influence on propensity to use ecofriendly gardening practices (Ecopract). As with several prior studies’ findings (Behe et al., 2010; Yue et al., 2010, 2016), higher educational attainment (College) was found to have a positive influence on Ecopract. Higher annual expenditures on gardening supplies (Ann Expend) were positively correlated with the latent variable Ecopract. This result is similar to findings regarding income by Khachatryan et al. (2017), Rihn et al. (2016), and Yue et al. (2010). Self-perceived lack of environmental knowledge (LackEnvKnow) was negatively correlated with the latent variable Ecopract. Also, if the gardener more strongly agrees that the current generation has an environmental stewardship responsibility to future generations (RespFutGen) and they like to grow their own food (GrowFood), they are more likely to exhibit an affinity toward adopting ecofriendly practices (Ecopract). These findings regarding environmental knowledge and concern support past studies’ findings suggesting a positive influence of environmental concern on ecofriendly gardening decisions (Campbell et al., 2020; Hugie et al., 2012; Kiesling and Manning, 2010). With respect to the effects of information sources, the coefficients on Other Gardeners, TV Magazines, and Extension/Master Gardener are each positive and significant.

As for the effects of the causal variables on propensity to prefer the environmental attributes in gardening supplies, the estimated coefficient on AgeGT50 is positive and significant in the structural equation for Envattrib. This finding suggests older gardeners exhibit a greater propensity to prefer ecofriendly attributes promoting gardening supplies. This finding is similar to that by Yue et al. (2016). Although self-perceived lack of environmental knowledge, LackEnvKnow, negatively influences Envattrib, respondents who believe that current generations have a stewardship responsibility to future generations for the environment (RespFutGen), that home gardeners can affect the environment (HomeGardEnvir), and who like to grow their own food (GrowFood) are more likely to prefer environmental attributes in garden supplies. These findings imply that environmental knowledge and concern positively influence propensity to prefer environmental attributes in gardening supplies and are similar to those by Campbell et al. (2020), Hugie et al. (2012), and Kiesling and Manning (2010). Also, use of information from other gardeners (Other Gardeners) positively influences Envattrib, suggesting gardeners seek out information from other gardeners to make decisions about purchasing gardening supplies with environmental attributes.

Interesting patterns emerge when the significance of the demographic, expenditures, information, and attitude variables across Ecopract and Envattrib are compared. In general, a greater number of demographic and expenditure variables significantly influence practice use than attribute preference. Male gender, college graduate level of education, and greater gardening expenditures positively influenced practice use, but not preferences for attributes. One possible explanation is that products with environmental attributes may hold appeal across a relatively broad cross-section of outdoor home gardeners, whereas implementation of practices may appeal to a more specialized segment of outdoor home gardeners. An exception to this is gardener age, with home gardeners who are older than 50 years being more likely to prefer environmental attributes, but age older than 50 years having no significant effect on practice use. One possible explanation is that older gardeners may favor supporting environmentally friendly gardening through their purchases. Their environmental preferences for practices may be weighed against the labor required to implement them. However, further research would be needed to fully investigate this possibility.

Although information from other gardeners influences both practice use and preferences for attributes, other information sources, including television and magazines and extension or Master Gardener programs, significantly influence practice use but not preferences for garden supply attributes. These results could indicate that home gardeners look to these particular sources for information about “how-to” guides, instructions, and illustrations for implementing ecofriendly gardening practices. Also, the results could suggest outdoor home gardeners may use a wider set of information sources for advice on practices, but a more narrow set, other gardeners, for making environmentally friendly product choices. Gardeners may rely more heavily on “word-of-mouth” recommendations from other gardeners for making product choices. Furthermore, an interesting finding is that although attitudes about environmental knowledge and responsibility to future generations for the environment both positively influence Ecopract and Envattrib, this is not the case for agreement with home gardeners being able to influence the environment with their actions. Although the coefficient on HomeGardEnvir is not significant in the Ecopract structural equation, it is positive and significant in the equation for Envattrib. One possible interpretation of this result may be that home gardeners believe their actions collectively as gardening product purchasers may have more effect on the environment than if they individually use the specified ecofriendly practices. However, further research would be needed to fully explore this possibility.

The estimated MEs of causal variables (defined in Table 1) are shown in Table 3. The MEs are interpreted as the change in the probability of the indicator variable being “1” (of using a practice or preferring an attribute) with a one-unit change in the causal variable. For example, the ME of female gender (Female) on using rainwater collectors (Rainwater) is −0.029, hence female gardeners are 2.9% less likely to use this practice than male gardeners.

The marginal effect on age (AgeGT50) ranges from 3.4% for Decr Need Pestic to 5.8% for Recyclable Pkg. Gender (Female) has a −2.4% effect on Soil Testing, with MEs magnitudes ranging up to −6.2% for Recycling. Being a college graduate (College Grad) has positive marginal effects on soil testing and recycling, ranging in magnitude from 1.7% for Soil Testing to 4.3% for Recycling. The marginal effects of annual gardening supply expenditures (Ann Expend) suggest that for each additional $100 of gardening expenditures the probability of using Soil Testing increases by 1.6% and the probability of using Recycling increases by 4.1%.

With respect to the effects of the use of information sources, the use of information from other gardeners (Other Gardeners) has positive MEs both on the probabilities of using the practices and on the probability of preferring the attributes. For the attribute preferences, only the MEs on Other Gardeners are significant at the 90% confidence level or greater. The smallest ME is 3.4% for the probability of preferring the attribute Decr Need Pestic, whereas the largest ME for Other Gardeners is on the probability of using the practice Recycling at 9.1%.

The MEs of TV Magazines and Extension/Master Gardener are positive and significant for the probabilities of using all the evaluated ecofriendly practices. The smallest of these is for Extension/Master Gardener on the probability of using Soil Testing at 1.9%, and the largest is for using TV Magazines on Recycling at 10.9%.

The effects of lack of self-perceived environmental knowledge on the probabilities of using the six practices and preferences for the six attributes are presented in Table 3. The largest ME for LackEnvKnow is on the probability of using Recycling (−4.5%) and the smallest is on the probability of using Soil Testing (−1.8%) and for preferring Decr Need Pestic (−1.9%). The variables capturing gardener environmental attitudes have MEs ranging from 1.1% for GrowFood on the probability of using Soil Testing to 9.8% for RespFutGen on the probability of preferring Recyclable Pkg.

Probabilities of using the practices and preferring the attributes.

The predicted probabilities of the home gardener using the six practices are graphed over increasing propensity to use ecofriendly gardening practices (Ecopract) (Fig. 2). Recycling of gardening supplies packaging (Recycling) is adopted most widely across the outdoor home gardeners at both lower and higher levels of Ecopract. This practice’s adoption rises above 50% use at even moderate levels of Ecopract. At lower levels of Ecopract, the practice with the next highest probability of use is planting pollinator plants (Pollinators), but as the respondent propensity to use ecofriendly gardening practices rises, the probability of using composting (Composting) rises more quickly than planting pollinator plants. The practices for which probability is lowest are use of rainwater collectors (Rainwater) and soil testing (Soil Testing). These two practices reach only 50% adoption at high levels of Ecopract, suggesting their use is primarily reserved to those who are most intensely interested in using ecofriendly gardening practices.

Figure 3 displays the predicted probabilities of the home gardener citing the six attributes as important graphed over increasing propensity to prefer ecofriendly attributes in gardening supplies (Envirattrib). The attribute Decr Need Pestic is the most widely preferred across both lower and higher levels of Envirattrib. The probability of selecting Recyclable Pkg rises next, followed by Decr Need Water. The probabilities of selecting Prod Organically and Native Species follow similar trajectories as Envattrib rises. These two attributes have the lowest probabilities of being selected as important of the attributes examined, suggesting these attributes are more limited in their appeal, primarily to those most concerned about purchasing gardening supplies with ecofriendly attributes.

The effects of gardener market profiles on probabilities of using the practices and preferring the attributes.

The two profiles created for Ecopract, Profiles 1 and 2, are shown in the second and third columns of the upper half of Table 4 (see Table 1 for variable definitions). The two profiles created for Envattrib, Profiles 3 and 4, are shown in the fifth and sixth columns. Profile 1 (Profile 2), is for a gardener who is very unlikely (very likely) to use the ecofriendly gardening practices. The Profile 1 (Profile 2) outdoor gardener is 50 years old or younger (over 50), female (male), not a college graduate (a college graduate), does not reside in a rural area (rural residence) with $200 ($400) in annual gardening supplies expenditures, does not use (uses) social media or the Internet, other gardeners, TV or magazines, or extension or Master Gardener program information sources, disagrees (agrees) that she or he is knowledgeable about environmental issues, and in general agrees (disagrees) that they are concerned about environmental issues.

Profile 3 (Profile 4) is a gardener who is very unlikely (very likely) to prefer the ecofriendly attributes. The Profile 3 (Profile 4) gardener is aged 50 years or younger (aged older than 50 years), male (female), not a college graduate (college graduate), lives in a rural area (does not live in a rural area), and spends ≈$200 ($400) per year on gardening supplies. The Profile 3 (Profile 4) gardener uses (does not use) social media/Internet, other gardeners, and TV or magazines as information sources. The Profile 3 (Profile 4) gardener uses (does not use) information from extension/Master Gardener programs and is self-perceived to lack (to not lack) adequate knowledge to make environmental decisions. Profile 3 (Profile 4) gardeners disagree (agree) with the statements that we are responsible to future generations for the environment, home gardeners can impact the environment with their actions, and that they like to garden to grow foods.

The predicted probabilities of using each of the six gardening practices for the two profiles are in the second and third columns in the bottom half of Table 4. The probabilities of using the six practices increase when moving from Profile 1 to Profile 2. For example, the probability of a gardener using pollinators (Pollinators) increases from 9% for Profile 1 to 72% for Profile 2. The probability of recycling garden supplies packaging increases from 9% to 86%. The largest probability increase is projected when moving from Profile 1 to Profile 2 for recycling garden supplies packaging (Recycler) and composting (Composting), whereas the smallest probability increases are for rainwater collectors (Rainwater) and soil testing (Soil Testing). As shown in Table 4, for each practice, the 95% confidence intervals around the simulated probabilities for Profile 2 do not overlap with those of Profile 1, indicating the Profile 2 probabilities are statistically greater than Profile 1 probabilities.

The probabilities of preferring the six attributes across the two profiles created for Envattrib (Profile 3 and Profile 4) are shown at the bottom of Table 4 in the fifth and sixth columns. An example interpretation of these probabilities is that preferences for decreased need for fertilizer rise from 9% for Profile 3 to 95% for Profile 4. When moving from Profile 3 to Profile 4, the largest increase in probability is for Decr Need Fertil, increasing from 9% to 95%. The smallest increase is for Prod Organically, increasing from 11% to 66%. As shown in Table 4, for each attribute, the 95% confidence intervals around the simulated probabilities for Profile 4 do not overlap with those of Profile 3, indicating the Profile 4 probabilities are statistically greater than Profile 3 probabilities.