Measurement of attachment in human-animal interaction research

At least some of the variability in HAI research is thought to be due to the specific bonds between humans and their companion animals – which, when incorrectly measured, or unmeasured, could partially moderate outcomes (Rodriguez et al., 2021). There are many different measures of the HAB, including the Pet Attachment and Life-Impact Scale (PALS; Cromer and Barlow, 2013), the Short Attachment to Pets Scale (SAPS; Marsa-Sambola et al., 2016), and most notably, the Lexington Attachment to Pets Scale (LAPS; Johnson et al., 1992).

The LAPS serves as a good encapsulation of the measurement problem as it relates to the bond between a person and a companion animal. The LAPS is, as of writing this article in March 2024, the most cited measure of the HAB with 590 citations (per Google Scholar). It has been translated into Spanish (Ramírez et al., 2014), Italian (Testoni et al., 2017; Uccheddu et al., 2019; Riggio et al., 2021), German (Hielscher et al., 2019), Portuguese (De Albuquerque et al., 2023), and Japanese (Volsche et al., 2023). Many of the articles citing to the LAPS have used it as a measure of the HAB. However, there were a number of issues in the creation and initial validation of the LAPS.

The original study design of the LAPS (Johnson et al., 1992) was conducted via telephone survey in Fayette County, Kentucky. The sample was 94% White, with an age range of 18–83 years. Given that the HAB construct appears to vary across the lifespan (Poresky and Daniels, 1998; Zilcha-Mano et al., 2011; Hirschenhauser et al., 2017), and that the sample was not nationally representative, this initial validation may not generalize to the larger population. Sampling issues aside, some of the items in the LAPS have questionable face validity; for example, there is a conflation between attitudes and the HAB in Item C (“I believe that pets should have the same rights and privileges as family members”) and Item N (“Pets deserve as much respect as humans do”). The HAB construct is theoretically bound in terms of the relationship between a specific person and their companion animal, and ipso facto does not cover the concept of attitudes toward companion animals generally.

In addition to the face validity issue described above, the authors of the LAPS used flawed statistical methods in their initial validation. Specifically, they sought to investigate the factor structure of the LAPS but conducted a principal components analysis (PCA), which does not identify the factors underlying the covariation in a dataset but is instead intended to measure total variation for the purpose of reducing the number of variables in a dataset (Lilienfeld et al., 2015). Essentially, the researchers claimed to have found distinct factors, when in reality they reported components, which are not indicative of the underlying dimensions of the variables (Lilienfeld et al., 2015). This analytical error is not limited to the LAPS; it is featured in a number of measures of the HAB, such as the SAPS (Marsa-Sambola et al., 2016), a popular HAB measure for children and youth. There is additional value in conducting survey research that investigates the more granular, species-specific relationship between people and specific species (e.g., dogs, cats, bunnies). These measures would allow for complex analyses that could account for participants who have multiple species of companion animals. However, these measures often suffer from the same issues as mentioned above (e.g., Dwyer et al., 2006; Coleman et al., 2016; Howell et al., 2017). Furthermore, a dissertation utilizing a large – albeit predominantly White (91.5%) – sample, could not confirm the proposed factor structure of the LAPS due to poor model fit and concluded that the measure may not be capturing the HAB construct (Zaparanick, 2008). Taken together, these limitations indicate that the measures being used to measure the HAB could be improved upon in future research. It is also important to note that we are not advocating for ignoring or invalidating all previous research that has relied upon measures like the LAPS. Instead, it is our hope that the field can take note of the significant limitations of existing measures and use this study as a guidepost for improvement going forward.

Therefore, the present study was designed to explore the factor structure of common measures of the HAB. To better inform recommendations regarding the use of the SAPS and the LAPS, the research team chose to explore the factor structures of those measures. The aims of this study are to confirm the factor structures of the LAPS and SAPS and to propose suggested remedies in order to improve measurement of the HAB.

Participants were 589 undergraduate students who participated in the biannual “Mass Survey” of students at a large state university in West Texas. The students completed the measures online and received experimental credit for their time and effort.

Incoming undergraduate students at a large, public, Hispanic-Serving Institution in West Texas were invited to complete a series of measures including, but not limited to, the survey questions used in these analyses. There were originally 962 responses to the Mass Survey questions. Any participants missing responses to the pet questions were dropped from the analytic group, resulting in 734 responses. Then, any non-pet-owning participants were removed, resulting in a final analytic sample of 589. The study was approved by the university’s institutional review board.

Given that the LAPS and SAPS have received scant attention to their factor structure and underlying psychometric properties, analyses were conducted in an attempt to replicate the originally proposed components using confirmatory factor analysis (CFA). As mentioned above, CFA functions differently compared to other statistical techniques, such as PCA. For example, CFA assumes that both common (resulting from the factor) and unique variance exist for indicators. By accounting for this unique variation, CFA provides a more accurate estimate of the variation resulting from the common factor(s) therein (Kim, 2008). Further, though PCA is often used in place of CFA, it does not assume an underlying factor structure and may provide biased results if used for this aim (Kim, 2008; de Winter and Dodou, 2016). Analyses were conducted in Mplus version 8.10 (Muthén and Muthén, 2017). Pet-owning participants were partitioned into two distinct data sets using random assignment (SAPS n = 295; LAPS n = 294). Monte Carlo power analyses indicated that these sample sizes were appropriate for estimating the observed parameters (Muthén and Muthén, 2002).

Given that participants answered questionnaire items using a Likert scale, indicators were modeled using weighted least-squares mean and variance-adjusted (WLSMV) estimation (Muthén, 1993). Several model fit indices were used for each CFA, such as the chi-square, the comparative fit index (CFI), the Tucker-Lewis index (TLI), and the root mean square error of approximation (RMSEA). Adequate model fit was indicated for CFI and TLI values > 0.90 and an RMSEA value less than < 0.06 (Hu and Bentler, 1999). Further, we assessed the internal consistency of the SAPS and LAPS using alpha coefficients. We also examined multicollinearity by using the variance inflation factor (VIF) and tolerance using SPSS Version 29 (IBM Corporation, 2023). VIF values greater than 10 and less than 0.10 were used as benchmarks.

The demographic characteristics of the sample are included in Table 1. The demographic characteristics between SAPS and LAPS groups do not significantly differ on gender, trans identity, race/ethnicity (χ²[18] = 12.06, P = 0.844), or age (t = −1.75[451.79], P = 0.081). Both the SAPS (α = 0.793) and LAPS (α = 0.939) demonstrated acceptable internal reliability (Tavakol and Dennick, 2011). VIF values for both the SAPS (1.383–2.781) and LAPS (1.387–3.506) were within the acceptable range, suggesting little evidence of multicollinearity (Hair et al., 1998).

Estimates for the SAPS model resulted in acceptable CFI and TLI estimates above the cutoff of 0.90. However, the model also resulted in an estimated RMSEA above the acceptable cutoff of 0.06. Several mechanisms were targeted to improve model fit. First, factor loadings were examined to identify items that did not significantly load onto the factor (i.e., factor loadings less than 0.4; Matsunaga, 2010). However, all factor loadings were above this cutoff. The smallest factor loading was item seven (0.587). Second, modification indices were examined. Several indices were identified but were all similar in their possible reduction of the chi-square value. There is limited theoretical evidence that would warrant the application of modification indices or removal of an item. Thus, the model was not altered to improve its fit.

Regarding the LAPS model, like the SAPS model, acceptable CFI and TLI values were identified. However, the estimated RMSEA fell below the acceptable cutoff. Again, estimates were evaluated to identify weak factor loadings. All loadings were above 0.4, the smallest being item 15 (0.527). Further, estimated modification indices would have resulted in similar reductions to the chi-square value if incorporated. Limited theoretical evidence exists for removing an item or specifying a new model using identified modification indices. Therefore, we did not attempt to alter the model to improve model fit. Fit indices and factor loadings for the SAPS and LAPS are included in Tables 2–4.

The data used as the basis for these analyses were collected via convenience sampling of incoming students at a large state university in West Texas. Though data missingness was not a particular concern, it is possible that participants could have misunderstood some of the questions, as there were no study administrators in the immediate vicinity to assist, which is one limitation of the online survey format. Additionally, although this study draws from a more racially diverse sample than the original Johnson et al. (1992) article, our sample is still non-representative and is in fact less representative in terms of gender compared to the Johnson et al. (1992) article. It is possible that our sample is biased as a result of its demand characteristics, or also because of the context from which it is drawn; therefore, we would like to re-emphasize that this article is supportive of other research groups’ undergoing validation attempts of the LAPS and SAPS on nationally representative samples (as well as samples that are particularly relevant to their own theoretical conceptualization; i.e., local sub-populations), as construct validation is never complete (Lilienfeld et al., 2015) and it is possible that the validity of the LAPS and SAPS may bear out.

The authors have no financial or non-financial conflicts of interest to declare.

The study was approved by the university’s institutional review board. All relevant guidelines were followed by the research team.

The last author named is lead and corresponding author. EH and TH contributed in writing – original draft; EH, TH, AH, and JV performed writing – review and editing; EH and TH conceptualized the data; EH, TH, AH, and JV investigated the study; EH, TH, and JV carried out methodology; TH carried out formal analysis of the study; and JV carried out project administration.

No external funding was relied upon or received in support of this research project.

For privacy purposes, the data are stored in a password-protected file accessible only to the authors. A copy of the data may be requested by emailing the first author.