This sample was a subset of the data underlying the validation of the TUNA database sampled in 2021 (Hahn et al., 2024). This subsample contains participants who rated only one meat picture. This picture was randomly selected out of 130 (not) disassembled and (not) cooked meat pictures (see Figure 1). As such, the individual pictures were rated by multiple participants. Of the 130 meat images, 80 had at least one matched counterpart across the factor levels of (not) disassembled and/or (not) cooked. The remaining images depicted meats that typically exist in only one form, such as disassembled cold cuts, for which usually no not disassembled or not cooked equivalent exists. To keep the stimuli similar across factor levels, we included only the 80 matched images in our analyses, ensuring that perception differences can be attributed to the experimental variation of disassembling and cooking rather than uncontrolled stimulus properties.

Pictures were presented in the middle of the screen with the question and response option below the picture. Participants first answered dichotomous questions on familiarity and recognizability of the object displayed in the picture. Next, participants used a slider (ranging from 0 to 100 with endpoints labeled) to report their perceptions regarding emotional arousal, complexity, palatability, desire to eat the dish, anger, disgust, compassion, valence, as well as positivity, negativity, and felt ambivalence. The presentation of these questions was randomized, with the exception that valence always preceded positivity and negativity, followed by felt ambivalence (see Figure S1 in the supplemental material for a schematic overview of the survey).

Positivity and negativity were used to calculate the similarity intensity model (SIM) index using the following formula: (Positivity + Negativity)/2 - |Positivity - Negativity| (Thompson et al., 1995) to indicate potential ambivalence (index range between -50 to 200). For ease of interpretation, we added 50 to the SIM index so that it ranged from 0 to 250, where 0 refers to no conflict in evaluations and 250 maximum conflict in evaluations. Table 1 provides an overview of variables included in our analysis. Zero-order correlations between all analyzed variables can be found in Table S1 in the supplemental materials.

A total of 2,060 participants were recruited as a representative sample from the U.S. via Prolific. Due to incorrect responses to one of the attention check items, 17 participants were excluded. Additionally, because we only wanted meat eaters in our analysis, we excluded 132 participants who indicated that they followed a vegan or vegetarian diet. To ensure that pictures were comparable across factor levels, only participants who rated pictures that had a match were included, resulting in a final sample of N = 1,189 (M_age = 44.56; SD = 16.08, range = 18–84 years). Figure S2 in the supplemental material presents a CONSORT-style diagram illustrating how the final sample was derived. Detailed demographic information for the final sample can be found in Table 2.

All analyses were conducted in R (R Core Team, 2013). We tested the direct effects of disassembling and cooking on the perception of meat dishes. To do so, we used separate maximum likelihood multi-level models (mixed-effect models) predicting the continuous variables outlined in Table 1. The models included disassembling (0: not disassembled, 1: disassembled) and cooking (0: not cooked, 1: cooked), and their interaction as predictors. To account for the non-independence of ratings due to the multiple evaluations of the same meat image across participants, we treated the specific meat dish (i.e., the images) as a level-two grouping variable with a random intercept. To account for potential confounding factors, we included demographic variables commonly associated with meat-related perceptions—namely gender, age, political orientation, eating style, and meat consumption—as covariates in the model (Buttlar et al., 2024; Kubberød, Ueland, Tronstad, et al., 2002; Michel et al., 2021; Pfeiler & Egloff, 2018). To test for the significance of the predictor variables, we tested these models via analyses of variance applying the Satterthwaite's method. As we were interested in the individual effects of disassembling and cooking on these outcome variables (i.e., individual testing as compared to disjunction and conjunction testing), we did not correct for multiple comparisons (see Rubin, 2021).

While multi-level models shed light on how cooking and disassembling affect the perception of meat dishes on individual variables, we also wanted to examine whether cooking and disassembling affect how these perceptions are interrelated. Psychological network analysis is a data-driven approach that allows us to investigate how cooking and disassembling affect the associations between meat-related perceptions (Borsboom et al., 2021). This type of analysis provides a visually insightful representation of how different meat-related perceptions are interrelated. In psychological networks, observed variables are depicted as circles (i.e., nodes) and statistical relationships between these variables as lines (i.e., edges). Such edges are estimated with data and represent partial relations, that is, relations between variables after controlling for the effects of other variables in the network. As such, network analysis can provide insight into unique relations between variables that cannot be explained by other variables in the network.

The networks in this study were estimated with the mgm model (mixed graphical models; Haslbeck & Waldorp, 2020), as this model allows for including continuous and categorical variables in the same network. This model estimates edges using regularized nodewise regression, meaning it calculates the relationships between each pair of nodes through regression analyses while controlling for all other nodes in the network and then applies regularization to remove weak edges in order to reduce the probability of false positive edges. Since the data are cross-sectional, edges represent conditional associations and not causal relations. An edge is included in the network if at least one of the two possible directions of relationships (from node A to B or vice versa) between nodes is selected. The interpretation of edges in a mgm network depends on the type of nodes involved: edges between two continuous nodes can be interpreted as partial correlations, and edges involving categorical nodes can be interpreted as (averaged) regression coefficients (Burger et al., 2023). Groups of nodes with a similar color represent communities of highly interrelated nodes as detected by a community finding algorithm.

Moreover, network analysis allows for comparing networks of different groups to examine whether meaningful differences can be observed, which allows for comparing the psychological networks of different conditions (i.e., cooked and not cooked; disassembled and not disassembled). Specifically, differences in network connectivity (i.e., the number and weights of edges between nodes) between networks from different groups are tested using the global strength metric, and differences in network structure (i.e., differences in edges between nodes) are tested using network invariance tests. Technical specifications of the R packages and their settings are provided in the supplemental materials S4.

Cooking affected the desire to eat meat dishes directly (F(1, 73.64) = 9.28, p = .003, ω_p² = .10), suggesting that cooking increased the desire to eat meat dishes (see Figure 2). Cooking also positively predicted valence (F(1, 76.06) = 4.17, p = .045, ω_p² = .04), and negatively predicted negativity (F(1, 75.92) = 4.71, p = .033, ω_p² = .05), felt ambivalence (F(1, 77.64) = 9.84, p = .002, ω_p² = .10), anger (F(1, 75.30) = 11.11, p = .001, ω_p² = .12), disgust (F(1, 76.09) = 10.87, p = .002, ω_p² = .11), and compassion (F(1, 1178) = 9.12, p = .003, ω_p² = .01).¹1

The multi-level model for compassion indicated a singular fit, suggesting negligible variance attributable to the grouping factor “meat picture”. Consequently, we simplified the model to a standard linear regression without random effects, being justified by the lack of evidence for meaningful variation in the grouping factor meat picture.

In contrast, disassembling did not affect the desire to eat meat (F(1, 73.90) = 0.29, p = .593). However, disassembling negatively predicted more general emotional arousal (F(1, 77.64) = 12.89, p = .001, ω_p² = .13) while also negatively predicting compassion (F(1, 1178) = 25.87, p < .001, ω_p² = .02). This suggests that disassembled meat elicited less general emotional arousal and compassion (see Figure 2B). There were no significant interactions between disassembling and cooking (all ps > .12). Statistical parameters for these models and estimated marginal means can be found in Tables S2 and S3 in the supplemental materials.

The network of the total sample (N = 1,189) is depicted in Figure 3 (see Table A1 in the Appendix for all edges in the network and Table S4 for results of edge accuracy and difference analyses). The orange community (#1) mainly encompassed emotions and consisted of compassion, emotional arousal, anger, disgust, felt ambivalence, and potential ambivalence. The blue community (#2) encompassed aspects of appetite, consisting of the desire to eat the meat dish and attitude aspects (palatability, familiarity, and valence). This distinction in communities suggests that the desire to eat the meat dish—as a proxy for behavioral intentions—is more strongly related to variables related to appetite than to variables related to emotions.

Specifically, the desire to eat meat dishes was positively related to valence (edge weight = .32), palatability (edge weight = .21), emotional arousal (edge weight = .20), familiarity (edge weight = .09), and compassion (edge weight = .08). These results indicate that, on average, higher scores on these variables tend to co-occur with a stronger desire to eat the meat dish and vice versa. Conversely, the desire to eat meat dishes was negatively related to disgust (edge weight = -.08) and potential ambivalence (edge weight = -.08). This indicates that, on average, more disgust and potential ambivalence tends to co-occur with less desire to eat the meat dish and vice versa. Taken together, these results suggest that higher disgust and conflict experiences (in the form of potential ambivalence) might, in fact, be related to reduced meat consumption, while higher valence, palatability, emotional arousal, and compassion are presumably related to higher meat consumption.

Network comparison tests showed no significant difference in the global strength of the networks between the cooked and not cooked (S = 0.26, p = .52) or the disassembled and not disassembled (S = 0.21, p = .58) networks. Results from a network invariance test were also not significant (cooked and not cooked: M = 0.25, p = .12; disassembled and not disassembled: M = 0.18, p = .54). In sum, the networks of relations between perceptions related to meat do not seem to be affected by the degree of disassembling and cooking of the different types of meat dishes.

Our research offers nuanced insights into how food preparation influences people's perceptions of meat. In line with the dissociation hypothesis (Benningstad & Kunst, 2020), we replicated findings that show that food preparation reduces disgust and compassion and increases the desire to eat the meat dish (Kubberød, Dingstad, et al., 2006; Kunst & Hohle, 2016; Kunst & Palacios Haugestad, 2018) and additionally showed, in line with the theory, that food preparation indeed affects people’s attitudes towards meat and conflict experiences (Bastian & Loughnan, 2017; Buttlar & Pauer, 2024). While we assume that these effects arise because people dissociate meat from its animal origin due to food preparation, the dataset on which we based our secondary analysis did not directly measure this dissociation process. To corroborate our findings, future work that builds on the present research should incorporate this measure and test the meat-animal dissociation as a mechanism for these effects more directly.

Surprisingly, emotional arousal and compassion were positively related to the desire to eat the meat dish in both zero-order correlations and network analysis, which indicates that higher emotional arousal and compassion tend to co-occur with a stronger desire to eat meat dishes and vice versa. These unexpected results might stem from the fact that these variables were assessed using rather general wording (see Table 1). That is, we did not specify whether the emotional arousal referred to positive or negative arousal; and the compassion item was without explicit reference to whom the compassion was directed. As such, these items might be interpreted heterogeneously across participants. For instance, emotional arousal may also be understood as the positive excitement related to eating a (tasty) meat dish, while compassion may not only incorporate compassionate feelings towards (slaughtered) animals but also reflect sympathy for others, as in the context of a family dinner. Future research should thus consider measuring emotional arousal as negative emotional arousal to reflect the aversive conflict, which is hypothesized to co-occur with cognitive conflict (Buttlar et al., 2025); it should also measure compassion directly in regard to animals, as compassion is especially suggested to decrease meat consumption if it is related to animal suffering (Benningstad & Kunst, 2020).

While the effects of our experimental variation of disassembling and cooking on meat-related perceptions were the main focus of our analysis, some limitations of the cross-sectional investigations on the relation between meat-related perceptions should be acknowledged. First, the relationships between variables are undirected. That is, higher meat-related disgust, for example, might have led to a lower desire to eat meat dishes; however, a lower desire to eat meat dishes might also have led to increased disgust ratings. Thus, directionality, such that disgust reduces meat consumption, may only be inferred from previous literature (e.g., Becker et al., 2022), but cannot be made from the data itself. Moreover, as relations between nodes are conditional on other nodes, the inclusion of nodes in a network might affect the results. Thus, our results depend on the perceptions of meat that were assessed in our study. While we think that the networks in this study provide a holistic overview of relevant meat-related perceptions and their relation to behavioral intentions, future research might uncover additional variables that shape people’s perceptions of meat as a function of disassembling and cooking and its effects on meat consumption.

In our study, we only visually presented the meat dishes to participants. Such visual impressions of meat are important, for instance, when people decide whether to buy packaged meat in the supermarket. In these situations, our findings may shed light on how cooking and disassembling affect the visual appearance of meat and, thereby, consumer perceptions and decisions. However, how meat is perceived is often also influenced by its smell, flavor, and texture (Font-i-Furnols & Guerrero, 2014; Kubberød, Ueland, Tronstad, et al., 2002; Michel et al., 2021; Ruby et al., 2016). This limits the generalizability of our findings, as it may be less applicable in situations where the desire to eat meat is not only elicited by the visual appearance but also by the smell of meat, for instance, when observing a meat dish on a neighboring table in a restaurant before deciding what to order. Although it is more challenging to test how these variables influence meat-related perceptions, studies would provide a more comprehensive perspective on how meat preparation affects people’s perception of meat.

In this vein, it is worth noting that disassembling did not directly affect the desire to eat meat, as would be expected based on the findings of Kunst and Hohle (2016). In their studies showing these effects (Studies 2a and 2b), the authors manipulated disassembling by presenting participants with a roasted pig with or without its head. In contrast, despite depicting a broad variety of meat dishes, none of the disassembled dishes presented in the present research depicted an animal with its head. This might imply that only very apparent reminders of the animal origin might affect people’s desire to eat meat, and less apparent disassembling does not suffice to do so (e.g., from distinctive cuts to minced meat). This effect might be partially driven by people’s familiarity with meat because they are usually less familiar with meat dishes consisting of whole animals (Possidónio et al., 2022). To this end, experimental research has shown that meat disassembling, which hides its animal origin, affects people’s appetite for visual depictions of meat mostly when meat dishes are unfamiliar but not when they are familiar to them (Possidónio et al., 2022). In line with this explanation, 78% of the participants in our research indicated that they were familiar with the dish. This familiarity might consequently have assuaged the effects of disassembling on the desire to eat meat. Thus, while we already depicted a broad variety of meat dishes, the current investigation might be extended by the addition of whole animals. This latter point might be especially important given the cultural variation in meat consumption.

While our data comprised omnivores drawn from a representative sample, this sample only represents omnivores from the U.S., and the effects of disassembling and cooking might differ in other cultural contexts. For instance, Kunst and Palacios Haugestad (2018) demonstrated that food disassembling helps people from Ecuador and the U.S. to dissociate from its animal origin, which decreased empathy and disgust but increased their willingness to eat meat in favor of a vegetarian alternative. Yet, these effects seem to be less strong the more often people are exposed to not disassembled meat, which varies across both cultures. As such, people from the U.S. seem to dissociate more than people from Ecuador. Thus, future research should try to uncover boundary conditions that moderate the effects of disassembling and cooking on the perception of meat dishes, especially in cross-cultural investigations.

How meat dishes are perceived is a matter of preparation. In this study, we show that food preparation decreases compassion, partially supporting the dissociation hypothesis (Benningstad & Kunst, 2020). These effects on compassion elicited by meat were present for different meat preparation steps, including disassembling and cooking. Yet, cooking and disassembling seem to have distinct effects, as disassembling mostly seems to affect general emotional arousal, whereas cooking seems to affect the desire to eat meat, meat-related attitudes, felt ambivalence, and specific emotions of disgust and anger. This way, the present investigation provides insights into how food preparation may serve as a passive conflict avoidance strategy and affect people’s desire to eat meat.