2.1. Survey development

The data for this study were collected through an online survey distributed via Qualtrics Panel Services (Qualtrics LLC, Provo, UT, USA). The survey was conducted in compliance with the university’s Institutional Review Board-approved protocol (IRB2022-0415). The survey sample was restricted to individuals eighteen years or older, who currently employ a LSP or who indicated they were considering hiring one. In addition to the choice-based conjoint section of the survey, the second section of the questionnaire collected demographic data, including the number of adults and children in the household, age (at the time of the survey), gender, race, education level, state of residence, region, and self-reported income level.

To ensure data quality, several exclusion criteria were applied. Automated Qualtrics protections removed duplicate IP addresses and automated (bot) responses. Two attention-check questions were embedded in the survey; responses failing either were excluded. Responses completed in less than one-third of the median completion time were flagged as speeders and removed. Respondents who were not eighteen years of age at the time of the survey, or who indicated they did not currently have, or were seeking a LSP, were excluded. Responses that failed any of these quality checks were not included in the final data set.

To measure WTP for scouting programs, a choice-based conjoint analysis was employed. Choice-based conjoint analysis, sometimes referred to as trade-off analysis, involves presenting survey respondents with a choice set of products that vary in attribute levels, forcing them to choose between products and “trade off” certain attributes for others. The resulting analyses yield part-worth values for each attribute, indicating the relative importance of a given attribute in the respondent’s decision to purchase (Behe et al. Reference Behe, Barton, Hall, Safley and Turner1999; Hopkins et al. Reference Hopkins, Hall, Arnold, Palma, Knuth and Pemberton2022) The part-worth values can then be used to calculate consumer potential WTP, help construct the ideal product, and make predictions on what consumers might be willing to pay for a product with a given set of attributes based on their part-worth values (Behe et al. Reference Behe, Barton, Hall, Safley and Turner1999; Boyle et al. Reference Boyle, Holmes, Teisl and Roe2001; Hopkins et al. Reference Hopkins, Hall, Arnold, Palma, Knuth and Pemberton2022; Hugie et al. Reference Hugie, Yue and Watkins2012; Palma et al. Reference Palma, Wirth, Adams and Degner2010). Each of the eight choice sets presented to respondents included an explicit opt-out alternative, allowing respondents to reject all alternatives when none were acceptable. Data were analyzed in Stata (StataCorp. 2023. Stata Statistical Software: Release 18. College Station, TX: StataCorp LLC.).

2.2. Attribute identification

Four attributes were examined: price, the potential for reduction of pesticide applications, automatic incorporation of biological control releases, and consumer choice of alternative/non-chemical control measures (Table 1). Price had three levels: included in the price of regular program fees, 10% more in cost of regular fees, and 20% more in cost of regular fees. Participants were asked to report their current monthly payment for landscaping services. This amount was then carried forward by Qualtrics and displayed during the choice experiment section of the survey (Figure 1). For the econometric analysis, the mean of the reported monthly fees was calculated and used to define the “regular fee” baseline; price levels corresponding to 10% and 20% increases were computed relative to this mean and entered into the statistical models as dollar-valued price levels, allowing WTP estimates to be expressed directly in monetary terms (Jiang et al. Reference Jiang, Zipp and Jacobson2018).

Figure 1.

Example of the discrete choice experiment showing respondents attributes associated with a set of two landscaper scouting plans. The respondent’s reported monthly cost that they currently pay was forwarded and displayed with the product choices.

Table 1.

Scouting product attribute levels used to measure willingness to pay for a landscape service provider scouting program

The other attributes attempted to incorporate elements of IPM programs, specifically reducing pesticide spray potential, incorporating biological controls, and offering the option to include non-chemical treatment methods. Pesticide spray reduction potential had three levels: no reduction in sprays, one to three potential reductions in spray applications, and four or more potential reductions in spray applications. The incorporation of biological controls and the choice of non-chemical control alternatives both had two levels of yes or no (Table 1).

The experiment was designed as a 3 × 3 × 2 × 2 factorial, with 36 possible product combinations. To prevent potential survey fatigue, attributes and levels were entered into JMP (JMP® Pro 18.0.2, SAS Institute Inc., Cary, NC) design of choice experiments tool to produce eight product choice sets for presentation to survey respondents (Table 2) (Behe et al. Reference Behe, Barton, Hall, Safley and Turner1999; Boyle et al. Reference Boyle, Holmes, Teisl and Roe2001; Hopkins et al. Reference Hopkins, Hall, Arnold, Palma, Knuth and Pemberton2022). The experiment was designed as a fractional factorial discrete choice experiment. Each respondent was presented with eight choice sets with varying product profiles; an opt-out option was included within each choice set (Table 2, Figure 1). The order of choice sets and alternatives within sets was randomized for each respondent to minimize order effects. To ensure response quality and minimize potential bias in model estimation, observations in which respondents selected the non-dominant alternative (specifically, the second option in choice set 3 and the first option in choice set 4 were excluded before estimating the conditional and mixed logit models (Bliemer et al. Reference Bliemer, Rose and Chorus2017). Data were analyzed in Stata.

Table 2.

Conjoint analysis choice sets used to measure willingness to pay for a landscape service provider scouting program

2.3. Statistical methods and models

To answer the first research question, which was to estimate consumer WTP for a scouting service offered by a LSP, we relied on choice-based conjoint analysis models grounded in McFadden’s random utility theory (Hu et al. Reference Hu, Woods and Bastin2009; McFadden Reference McFadden1973). Generally, random utility theory states that every alternative from which a consumer chooses has value or “utility” (Hu et al. Reference Hu, Woods and Bastin2009; McFadden Reference McFadden1973). This utility has two main parts: the deterministic term (can be observed) and the random term (cannot be observed). The random utility model is expressed as:

(1) ${U_{ijt}} = {X_{ijt}}B + {e_{jt}}$

Where individual i faces a choice of alternative j (scouting plan profile) in the t-th choice set, with attribute levels represented by X _ij. The β is the unknown parameter to be estimated, and e _jt is the error term.

In discrete choice experiments, WTP represents the monetary value that respondents assign to specific attributes of a good or service and is typically derived from the estimated coefficients of a choice model, such as the multinomial logit or mixed logit model. The WTP for a given non-monetary attribute is calculated as the negative ratio of that attribute’s coefficient to the price coefficient:

(2) $WTP = - \;{{\beta \;attribute} \over {\;\beta \;price}}$

Equation (2) quantifies the amount respondents are willing to pay for a one-unit change in the attribute, holding all other factors constant. A positive WTP indicates that the attribute increases utility and is valued by respondents, while a negative WTP suggests the attribute is undesirable (Behe et al. Reference Behe, Campbell, Hall, Kachatryan, Dennis and Yue2013; Campbell et al. Reference Campbell, Kachatryan, Behe, Dennis and Hall2015; Mason et al. Reference Mason, Starman, Lineberger and Behe2008; McFadden Reference McFadden1973)

Two of the most common models used to analyze choice-based experiments are conditional logit and mixed logit regression (mixed logit) models (Behe et al. Reference Behe, Barton, Hall, Safley and Turner1999; Boyle et al. Reference Boyle, Holmes, Teisl and Roe2001; Hopkins et al. Reference Hopkins, Hall, Arnold, Palma, Knuth and Pemberton2022; Hu et al. Reference Hu, Woods and Bastin2009; McFadden Reference McFadden1973). Conditional logit models apply McFadden’s theory and assume all individuals have equal preferences and are more restricted by the independence of irrelevant alternatives (IIA) assumption (Train Reference Train2003).

Mixed logit models enable the expansion of McFadden’s framework by allowing the coefficients or preferences to vary across individuals, capturing the unobserved heterogeneity of individuals and relaxing the IIA assumption (Behe et al. Reference Behe, Barton, Hall, Safley and Turner1999; Boyle et al. Reference Boyle, Holmes, Teisl and Roe2001; Hopkins et al. Reference Hopkins, Hall, Arnold, Palma, Knuth and Pemberton2022; Hu et al. Reference Hu, Woods and Bastin2009; McFadden Reference McFadden1973). Estimating conditional and mixed logit models helps conceptualize random utility theory by offering different levels of complexity based on how individual preferences are treated (Behe et al. Reference Behe, Barton, Hall, Safley and Turner1999; Boyle et al. Reference Boyle, Holmes, Teisl and Roe2001; Hopkins et al. Reference Hopkins, Hall, Arnold, Palma, Knuth and Pemberton2022; Hu et al. Reference Hu, Woods and Bastin2009; McFadden Reference McFadden1973). In the mixed logit model, the coefficients for spray reduction attributes, incorporation of biological control, and inclusion of non-chemical controls were specified as normally distributed random parameters, while the price coefficient was fixed. Random parameters were assumed to be independent. The estimated means and standard deviations of these random coefficients are reported in Table 4.

Conditional logit models assume homogeneous preferences and IIA. Mixed logit models relax these assumptions, allowing coefficients to vary randomly across respondents and capturing unobserved preference heterogeneity. Both models were estimated to provide complementary insights: conditional logit serves as a benchmark, whereas mixed logit offers greater flexibility. Consistent signs and magnitudes of coefficients across models enhance confidence in the robustness of findings, while improvements in log-likelihood and Akaike information criterion (AIC) for the mixed logit model confirm the presence of preference heterogeneity. Consequently, mixed logit estimates were used as the basis for WTP calculations (King et al. Reference King, Palma, Behe, Fernandez, Sage, Hall and Arnold2015; McFadden Reference McFadden1973; Roe et al. Reference Roe, Boyle and Teisl1996).

In addition to the choice experiment, respondents were asked a supplementary question regarding their likelihood of purchasing an IPM program from a landscaper. This item was designed to provide additional context on attitudes toward IPM and was not a primary focus of the study. Responses were collected using a four-point, forced-choice scale (very unlikely, somewhat unlikely, somewhat likely, very likely) to avoid a neutral midpoint and encourage directional responses (Wetzel et al. Reference Wetzel, Frick and Brown2020). Resulting in a multinomial logistic regression model with the response levels being the dependent variable and the same respondent demographics as the previous model being the independent variables. To answer the second research question, which is what services/attributes related to a scouting program would be lucrative for LSP to adopt, respondents were asked if they would purchase a biological control application and if they would purchase an IPM program offered by a LSP. The biological control response was binary (yes or no), and responses were analyzed using a binomial logistic regression model to estimate purchase likelihood. The dependent variable was the purchase of a biological control treatment (yes/no), with respondent demographics (household income, years of education, number of hours spent gardening, household size, and area of residence) as the independent variables.