Does Going Green Feel Good in Russia: Implicit Measurements With Visual Stimuli

Pro-environmental behaviour is of profound significance in shaping pathways towards sustainability. However, the approach to understanding pro-environmental behaviour in many spheres and cultures remains fragmented, and is often based on poor data. For a country as large as Russia, we have very little knowledge about everyday pro-environmental behaviour (Graves et al., 2019), its determinants (Sautkina et al., 2021), or even whether people view environmentalism positively. The latter relates to the existence of environmental attitudes, which can be defined as both the intensity of positive or negative affect about the general environmental domain and as a hierarchical attitude system that connects and organises more specific attitudes about a range of pro-environmental behaviours (Cruz & Manata, 2020). Investigating the environmental attitudes of Russians is important, as Russian environmental culture is still emerging (Valko, 2021; see also Appendix A), and this can provide some insights into how they view environmental issues and how they may respond to environmental policies.

There are still few articles in English on environmental culture and aspects of pro-environmental behaviour in Russia (Sautkina et al., 2022). These articles mostly focus on theoretical concepts, strategic goals (e.g., Shutaleva et al., 2020), legislation (e.g., Gladun & Zakharova, 2020), education (e.g., Ali & Anufriev, 2020), etc. To the best of our knowledge, only two studies have been published in English that have empirically explored the environmental, social, and political determinants of pro-environmental behaviour (see Sautkina et al., 2021) and environmental attitudes in Russia (see Valko, 2021), but they are based on self-reported data.

The former study revealed some differences from Western environmental cultures. For instance, social pro-environmental behaviour, in addition to integrated motivation, is positively explained by intrinsic motivation; populism is a positive predictor of resource conservation, but at the same time, resource conservation is rarely identified as pro-environmental behaviour itself; the relationships between environmental knowledge and environmental and psychosocial variables appear to be mostly negative; climate-relevant behaviour is negatively predicted by some socio-economic variables (e.g., income) (Sautkina et al., 2021). It is also reported that environmental concern is an appropriate predictor of pro-environmental behaviour in Russia. Environmental concern is a high-order attitude that relates to values and beliefs and can easily be linked to a positive perception of the environment (Schultz et al., 2004; Venhoeven et al., 2020).

The latter study suggests that people in Russia usually overestimate their own tendency towards pro-environmental behaviour, even under some affective interventions (Valko, 2021). This may be sufficient evidence of self-reported bias.

Recent articles have also emphasised the lack of adapted tools and scales to study pro-environmental behaviour in Russia itself and environmental attitudes (Ivanova et al., 2020), and the available data is mostly self-reported (Sautkina et al., 2022).

To address this complex gap, as an initial step, we adapted common indirect measures and a well-known scale of environmental concern that allows us to systematically investigate implicit and explicit attitudes. In the next part, we want to briefly present the current discussion about common approaches in this case.

The common discrepancy between what people say about global sustainability and what they do about the environment is explained by the influence of implicit environmental attitudes. There is an active debate about the implicit and explicit dimensions related to how attitudes are evaluated, namely whether this evaluation relies on a self-reported evaluation of the object or is derived from overt behaviour that excludes such self-reporting (e.g., Cruz & Manata, 2020; De Houwer, 2006; De Houwer & Moors, 2010). The vast majority of empirical studies examining attitudes towards sustainability and pro-environmental behaviour use explicit and self-reported measures (e.g., Taufik et al., 2016; van der Linden, 2018), others mostly use qualitative analysis (e.g., Meenar et al., 2022) or even autoethnographic methods (e.g., Halstead et al., 2021). The limitation of self-report measures in environmental psychology is routinely discussed in many papers (e.g., Kennedy et al., 2015; Milfont, 2009; Steg & Vlek, 2009). Self-reports of behaviour may be distorted by social desirability, consistency biases, participants’ inability to accurately recall their behaviour, and some individual differences in knowledge and the interpretation of a questionnaire (e.g., Gifford, 2014; Kormos & Gifford, 2014).

Certainly, the way in which such attitudes are measured can lead to contradictory results (Thomas & Walker, 2015; Venhoeven et al., 2020). This relates to the hierarchical structure of attitudes: they are connected to one another; higher-order attitudes are broad and abstract, and become progressively more specific and concrete as one moves down the hierarchy (Cruz & Manata, 2020).

A variety of research have implemented different techniques in order to implicitly investigate affective attitudes and their relationship with self-report and explicit measures across a range of domains. There are three commonly used techniques: Implicit Association Test (IAT, see Greenwald et al., 1998), Evaluative Priming Task (EPT), and Affect Misattribution Procedure (AMP). The last two of them are based on affective priming (AP), a psychological phenomenon in which exposure to a stimulus influences an individual’s subsequent evaluation of a target object.

The EPT is a well-known priming-based measure that has been used to investigate the automatic activation of attitudes (Fazio et al., 1986). It exploits the idea that attitudes are represented in memory as object-evaluation associations of varying strength (see Fazio, 2007). A central implication of this idea is that encountering an attitude object may automatically activate its associated evaluation via spreading activation to the extent that the associative link between the two is sufficiently strong (Gawronski et al., 2020).

On a typical trial of the EPT, participants are briefly presented with a prime stimulus of unknown affective valence, which is followed by a positive or negative word as a target stimulus (Fazio et al., 1995). The participant’s task is then to categorise the target stimulus as positive or negative as quickly as possible by pressing one of two designated keys. The basic idea underlying the EPT is that quick and accurate responses to the target stimuli should be facilitated when the prime stimulus elicits a compatible evaluative response with the valence of the prime stimulus. In contrast, quick and accurate responses to the target stimulus should be impaired when the prime stimulus elicits an incompatible evaluative response (Koppehele-Gossel et al., 2020).

Since then, researchers using the EPT have relied on a wide range of procedures to minimise noise in evaluative priming data (for a meta-analysis, see Herring et al., 2013). For example, it has been shown that cutting off 300-1000 ms in laboratory experiments can help to reduce the likelihood of false-positive/negative results due to outliers, too fast and too slow responses, etc. (Koppehele-Gossel et al., 2020). Critically, participants should be instructed to ignore the primes and not to let the primes influence their judgments of the target.

Cognitive research suggests that affective priming effects might be effectively detected in different tasks: categorization, classification, naming, rating tasks (Appel et al., 2021) or Stroop task (Damen et al., 2018). Theoretically, if the task or a target stimulus to be evaluated is compatible (congruent) with the prime stimulus, response times (in the evaluation phase) tend to be relatively shorter than for an incongruent pair. Thus, in order to test affective response to an environmentally related prime stimulus accompanied by a positive target stimulus, it is expected to elicit equivalent response time as for congruent pairs, and significantly different from those for negative targets.

However, the meta-analysis showed that in some cases the EPT tends to show rather low estimates of reliability (for instance, with renewal effects the EPT Cronbach’s α values between .00 and .55; see Gawronski & De Houwer, 2014). In these cases, studies also utilised the AMP, which has higher reliability (Cronbach’s α values between .70 and .90; see Gawronski & De Houwer, 2014).

The AMP was developed by Payne et al. (2005) as an implicit measure technique that is based on individuals’ tendency to make erroneous attributions of diverse phenomena. Originally, each trial on the AMP begins with a prime stimulus presented briefly, but not subliminally (Payne et al., 2005). The prime is followed by a target object (e.g., a pictograph) that is intended to be ambiguous or neutral with regard to the judgement made about it. When presenting an ambiguous stimulus, the subject tends to confer attributes based on previously developed mental representations or association, which are not always an accurate reflection of reality (Gawronski & Ye, 2014). Following the target object, a visual mask is presented to prevent subjects from inspecting the target for too long, or the target object can be simply hidden. Subjects eventually respond by making a judgement about the target object. In some conditions, subjects are warned not to be influenced by the primes. Theoretically, the target object elicits a more positive response if the prime stimulus is more pleasant, or vice versa. Therefore, for environmentally related prime objects, it is expected to elicit equal reported valence as for positive ones, and significantly different than for negative ones.

On most AMP studies, binary response scales have been used in the evaluation phase (e.g., pleasant/unpleasant; Payne et al., 2005), but Likert-type scales have also been used successfully (e.g., Payne et al., 2013; Pryor et al., 2012). Combining the logic of projective tests with advances in priming research, the AMP is sensitive to normatively favourable and unfavourable evaluations, and the misattribution effect is strong at both fast and slow presentation rates (Payne et al., 2005). The advantages of the AMP include large effect sizes, high reliability, ease of use, and resistance to correction attempts (Payne & Lundberg, 2014).

Thus, implicit measures have demonstrated an appropriate effect size, power value, and reliability (Sánchez et al., 2016); they are also suitable for assessing attitudes using visual stimuli (Gawronski & De Houwer, 2014; Pérez, 2013). Detecting positive attitudes towards environmentally related visual stimuli is sometimes effective regardless of whether the instrument is implicit or explicit: EPT, AMP (Hietanen et al., 2007; Hietanen & Korpela, 2004; Korpela et al., 2002; Sánchez et al., 2016), and those with explicit measures (Schultz & Tabanico, 2007; Venhoeven et al., 2020).

Therefore, we decided to use the AMP as the main implicit measure, and to adapt the technique of affective priming as a certain modification of the EPT.

A large number of scales measuring environmental attitudes, and environmental concern in particular, have been developed, and it can be challenging to choose a priori which one to use (Calbi et al., 2017; Somerwill & Wehn, 2022; Sparks et al., 2022). The environmental attitude scale by Wesley Schultz is a very simple 12-item questionnaire that is related to the value-basis theory for environmental attitudes and represents a three-factor structure (see, Appendix B; Schultz, 2001).

This questionnaire has been widely used as an explicit measure of environmental concern and reflects how people perceive and value themselves (egoistic attitude), others (altruistic attitude), and nature (biosphere attitude) in the context of environmental issues. It has also been used in studies identifying implicit connections to nature, as well as cognitive strategies associated with egoistic and biospheric attitudes (e.g., Cruz & Manata, 2020; Schultz et al., 2004). Since there is some evidence of successful adaptation of the scale to the Russian context (Sautkina et al., 2022), we decided to use it as a main explicit measurement. We also suggested a possible relationship between positive attitudes towards sustainable development and environmental concern as it is common for Western cultures (Venhoeven et al., 2020).

This study was preregistered in the Open Science Framework (OSF; Nosek & Lakens, 2014) prior to the analysis. Preregistration included pilot data, target sample size, the power analysis, the exclusion criteria, hypotheses, analytical design, stimulus materials, and online platform html/js-scripts: http://dx.doi.org/10.31219/osf.io/9ag46.

In our implementation we used a basic AMP scheme (Payne et al., 2005; Payne & Lundberg, 2014) with pre-rated¹ visual stimuli, a Likert-type 5-point scale (1 = extremely negative; 5 = extremely positive) and a simple three-stage user interface (Fig. 1) for each trial.

We showed each subject 16 trials, each trial consisting of a pair of randomly selected visual stimuli (see Materials). The subjects were instructed to ignore the primes and respond as quickly as possible whether the target image looked positive or negative. The trials were randomised but included primes of positive, negative, and environmentally related stimuli in fixed proportions (Appendix C). In total, we showed eight trials with randomised environmentally related primes, four trials with randomised pre-rated positive primes and four with randomised pre-rated negative primes.

Each trial lasted approximately 6 seconds (a two-second delay between trials) which is longer than in conventional laboratory experiments (from 100 to 2000 ms in the original paper; Payne et al., 2005), but suitable for an online-based survey (Calbi et al., 2017; Damen et al., 2018).

After all the trials were shown, the reported affective valence was aggregated across subjects and stimulus categories and then compared using the paired Wilcoxon signed-rank test². As a validity check for the intervention, we tested the correlation between the reported valence and the response time, indicating a non-automatic response. A negative response to a pre-rated positive stimulus (or vice versa) was treated as an invalid trial and the entire set of trials for that subject was excluded. As a common solution for the outlier issue, we excluded responses that were longer than 4 seconds. The description of the procedure and analysis has been preregistered, as have the exclusion criteria (https://osf.io/z9d4x).

In this procedure we adapted the basic idea of the AP/EPT. In our implementation, the categorization task was transformed with visual stimuli (as in Meissner & Rothermund, 2015) and a Likert-type 5-point scale (1 = extremely negative; 5 = extremely positive). The logic behind this procedure is the following: if the target stimulus is congruent with the prime stimulus, response time in the evaluation phase tends to be relatively shorter than for an incongruent stimulus pair, because changing the valence of the presented stimuli requires more cognitive control in the evaluation phase.

We used the same three-stage user interface (Fig. 2) and showed each subject 16 trials, each consisting of a pair of randomly selected visual stimuli (see Materials).

All the trials were randomised but covered prime + target combinations of positive, negative, and environmentally related stimuli in fixed proportions (Appendix C). In total, we showed eight trials: four with randomised environmentally related primes and pre-rated positive targets, and four with randomised pre-rated negative targets. Additional eight trials, consisting only of combinations of randomised positive and negative stimuli, were given to test whether congruent and incongruent trials worked as predicted.

Each trial lasted approximately 6 seconds (with two-second delay between trials) which is longer than in conventional laboratory experiments (e.g., Koppehele-Gossel et al., 2020), but suitable for an online-based survey (see Calbi et al., 2017; Damen et al., 2018; Meissner & Rothermund, 2015).

After all the trials were shown, response times were aggregated across subjects and stimulus categories, and then compared using the paired Wilcoxon signed-rank test. As a validity check for the intervention, we tested the difference in response times between pre-rated congruent and incongruent trials (true positive and true negative pairs). The subjects were instructed to ignore the primes and respond as quickly as possible, and we excluded the responses longer than 4 seconds. The description of the procedure and analysis has been preregistered, as have the exclusion criteria (https://osf.io/z9d4x).

We employed a 12-item scale by Wesley Schultz (Schultz, 2001; see Appendix B) and used structural equation modelling (SEM) analysis to examine possible relationships between implicit environmental attitudes and explicitly measured environmental concern (Fig. 3). SEM allows us to use the questionnaire responses as they are, without a separate confirmatory factor analysis, and obtain bootstrapped estimations with all available socio-demographic variables at once.

Basic stimulus materials were obtained from OASIS (Kurdi et al., 2017), an open-access dataset containing 900 colour images of a broad range of themes, as well as normative ratings of two affective dimensions: valence (i.e., the degree of positive or negative affective response evoked by the image) and arousal (i.e., the intensity of the affective response evoked by the image). We obtained a subset of images from OASIS with an average arousal in the activity-neutral “scene” category. The images in the upper quartile (Q3) of affective valence were marked as pre-rated “positive” stimuli (19 images), and the others in the lower quartile (Q1) were marked as pre-rated “negative” stimuli (16 images). A subset of the images with average valence was used as “neutral” stimuli. We also generated and mixed as neutral stimuli a number of monochromatic polygon images from a randomising service³ (16 images in total, see examples in an online repository: https://osf.io/ayk28/files/osfstorage).

A set of images that were used as environmentally related stimuli were obtained from public sources (Freepik, Flickr, etc.) under a non-commercial licence (or free to use) and selected by eight independent experts. We asked the experts to rate images that have a strong association to the sustainability domain and look related to pro-environmental behaviour. Their decisions were aggregated using a standard methodology based on the analytic hierarchy process (Cho, 2019). This resulted in the selection of 18 images consistently associated with each general pro-environmental behaviour category: recycling, eco-shopping, resource-saving, eco-mobility. These categories correspond to pro-environmental culture in Russia and are commonly used in such studies (e.g., Sautkina et al., 2021; Valko, 2021). The selected images were scaled to 250x200px, normalised and presented on a black screen. All the materials and protocols used are available in an open repository (https://osf.io/z9d4x, https://osf.io/ayk28/files/osfstorage).

We conducted two studies based on the same algorithm (Fig. 4) to ensure repeatability. The first, pilot study (study 1) was run from September to November 2021, and we gathered 183 responses from Russian citizens (M_age = 23.7, SD = 7.3; 28.3% male). The second study (study 2) was run from February to April 2022 after pre-registration and with minor technical revisions. The purpose of the second study was to reproduce previously found effects, as well as attempt to improve external validity.

A power analysis (using MorePower; Campbell & Thompson, 2012) showed that 104 subjects are sufficient to identify an effect size of .36 (the average Cohen’s d in social psychology; Lovakov & Agadullina, 2021) with a 6-modality within-subject response design and a power of .90. It was preregistered that we had to obtain nearly 200 responses (for possible incompleteness, incorrectness, and smaller effect sizes), so we finished study 2 when we had gathered 211 responses (M_age = 24.5, SD = 8.1; 26.7% male).

We implemented the measurement procedures as described above on a standalone website that is suitable for any device. All trials (16 AFP and 16 AMP, enough not to irritate the subject; Sánchez et al., 2016) were automatically shuffled and introduced to each subject in a randomised order. After all the trials were completed, the environmental concern questionnaire was shown. All the instructions and informed consent were given at the beginning. A sociodemographic variables questionnaire and debrief were placed at the end. Each subject was asked to participate only once. Ethical approval for these studies was obtained from the South Ural University of Technology Committee.

The same online survey was used for both studies, the link to which was distributed on university websites, independent resources and via Russian social networks without specific target preferences (vk.com, ok.ru). We consider this as a convenience sampling. The total sample size of the two studies is N = 394, covering 68 Russian cities; detailed characteristics of the data are presented in Appendix D.

The results of the AMP procedure are shown in Fig. 5. It demonstrates that in both studies the median reported valence for environmentally related primes (env.) is statistically equal⁴ to true positive ones (study 1: Mdn_env. = 3.1, 95%CI [3.0 – 3.2], Mdn_positive = 3.2, 95%CI [3.0 – 3.5], W = 3780.5, p = .267, r = .09; study 2: Mdn_env. = 3.1, 95%CI [3.0 – 3.2], Mdn_positive = 3.2, 95%CI [3.1 – 3.4], W = 5250.0, p = .155, r = .11) and significantly greater⁵ than that for negative ones (study 1: Mdn_negative = 2.8, 95%CI [2.6 – 2.9], W = 5990.0, p < .001, r = .32; study 2: Mdn_negative = 2.8, 95%CI [2.6 – 2.9], W = 8715.5, p < .001, r = .34).

As a validity check for the intervention, no significant correlation was found between the reported valence and the response times in env. trials (study 1: r_Spearman(138) = .05, 95%CI [-0.12 – 0.21], p = .555; study 2: r_Spearman(166) = .03, 95%CI [-0.12 – 0.18], p = .706), and there is predicted difference between true positive and true negative primes.

F-test was conducted to accurately compare if the subjects exhibited positive evaluations of the target after exposure to positive or env. stimuli than after exposure to negative stimuli. It was coded jointly as contrasts of (1,1,-2) and (1,-1,0) for exposure to env., positive and negative stimuli, respectively. There was a significant expected difference that was consistent with previous analyses: study 1: F_(2, 417) = 10.69, p < .001, d = .45; study 2: F_(2, 501) = 11.55, p < .001, d = .43.

The results of the AFP procedure are shown in Fig. 6. It demonstrates that in study 1 the median response time for environmentally related primes (env.) that are accompanied by positive targets is statistically equal⁶ to congruent pairs (study 1: Mdn_{env.+positive} = 1491.5, 95%CI [1444.4 – 1538.6], Mdn_congruent = 1510.1, 95%CI [1439.1 – 1581.2], W = 2909.5, p = .298, r = .10) and significantly lower⁷ than those with negative targets (study 1: Mdn_{env.+negative} = 1631.0, 95%CI [1550.9 – 1711.1], W = 1233.0, p < .001, r = .55). In study 2 we found a moderate equality between environmentally related primes with positive targets and congruent pairs (study 2: Mdn_{env.+positive} = 1541.5, 95%CI [1481.9 – 1601.1], Mdn_congruent = 1541.8, 95%CI [1472.2 – 1611.2], W = 3944.5, p = .039, r = .17). However, the response times for environmentally related primes with positive targets were still significantly lower⁸ than that for the ones with negative targets (study 2: Mdn_{env.+negative} = 1668.0, 95%CI [1586.9 – 1749.1], W = 1850.0, p < .001, r = .55).

As a validity check for the intervention, the difference between pre-rated congruent and incongruent pairs was also significant (study 1: Mdn_incongruent = 1581.6, 95%CI [1527.2 – 1636.0], W = 2397.0, p = .006, r = .26; study 2: Mdn_incongruent = 1570.2, 95%CI [1496.2 – 1644.3], W = 3904.0, p = .016, r = .20).

The design of the procedure enabled us to concurrently examine whether subjects respond more rapidly when the primes are positive or related to the environment and the target is also positive, rather than negative. It was coded jointly as contrasts of (1,1,-2)×(1,-1) and (1,-1,0)×(1,-1) for exposure to env., positive, and negative primes × positive, and negative targets, respectively. F-test revealed a significant expected difference that was consistent with previous analyses: F_(2, 680) = 4.33, p = .014, d = .23; study 2: F_(2, 836) = 4.99, p = .007, d = .22.

We performed SEM analysis based on complete questionnaires (see Fig. 4.). It has found no significant relationships between components of the environmental concern and affective valence in either study with or without sociodemographic controls (study 1 with sociodemographic variables: χ²(152, MLW, n = 125) = 249.95, p < .001, TLI = .86, CFI = .88, GFI = .75, RMSEA = .072; study 1 without sociodemographic variables: χ²(59, MLW, n = 125) = 159.90, p < .001, TLI = .80, CFI = .85, GFI = .78, RMSEA = .117; study 2 with sociodemographic variables: χ²(152, MLW, n = 151) = 214.59, p < .001, TLI = .95, CFI = .95, GFI = .86, RMSEA = .052; study 2 without sociodemographic variables: χ²(59, MLW, n = 151) = 130.52, p < .001, TLI = .92, CFI = .94, GFI = .90, RMSEA = .090; all the factor loadings and estimations are presented in Appendix E).

Despite the general discussion of the reliability of implicit measures (see Gawronski & Bodenhausen, 2006; Sánchez et al., 2016), one limitation of our study is the modifications we made to the original procedures. These modifications are not unique or untested in the literature, but we believe our validity checks, which produced predicted results with pre-rated stimuli, are sufficient to ensure their accuracy.

In addition to trials and stimuli randomization, we used several approaches to improve the robustness of our results: we used two different implicit measurement procedures (AMP and AFP), reported the intervention validity checks (correlation between response time and reported valence, as well as reported valence for pre-rated stimuli), employed a range of control variables in SEM analysis, and improved external validity and repeatability by conducting the experiment twice and online rather than in a laboratory.

Another limitation is the somewhat short Schultz’ questionnaire, which may be inadequate for the Russian context. The lack of appropriate methodological tools in Russia requires the adaptation of existing pro-environmental behaviour scales and the creation of more tailored ones for the national context (Ivanova et al., 2020). We have some evidence that this scale has been successfully used in the context of the Russian environmental culture (Sautkina et al., 2022).

A third limitation is the stimulus materials we used as related to pro-environmental behaviour, which primarily focuses on sustainability rather than certain pro-environmental practices. We would like to thank our experts who did their best to select a suitable set of images that are associated with pro-environmental behaviours and cover its four main categories.

Last but not least, our data sample is unfortunately unrepresentative of the Russian population, as it consists of young individuals and has a slightly imbalanced gender ratio. Future researchers could obtain a more representative and larger sample of data (including different characteristics such as household structure, religion, subjective norms, beliefs, values, etc.), use better measurement tools and materials, and conduct qualitative or mixed-method studies (e.g. Thomas et al., 2019) to gain deeper insights through interviews and observations.

To support reproducibility and open science principles, we have provided all the data, materials, and scripts in an open repository (https://osf.io/ayk28/files/osfstorage, https://osf.io/z9d4x). There is a Jupyter Notebook with full analysis that allows anyone to reproduce our calculations and graphs. The final published manuscript contains post-hoc contrast analysis and some little changes from the preregistered algorithms and descriptions due to comments from independent reviewers and editors.

It is challenging to answer the question what Russian people’s implicit environmental attitudes are like, as these attitudes may vary among individuals who engage in environmental initiatives and campaigns differently. However, based on our analysis, we can conclude that in the emerging pro-environmental culture of Russia, young people tend to have deep positive attitudes associated with sustainability and related behaviour. These results support the importance of an emotionally charged communication and promotion of pro-environmental behaviour, as early demonstrated by Valko (2021).

Our findings align with recent literature on positive attitudes towards pro-environmental activities and sustainability in Western cultures (e.g., van der Linden, 2018; Venhoeven et al., 2020). This suggests that cultural differences in Russia, if they exist (see Appendix A for additional remarks), may not strongly affect basic affective reactions in this context (Pröpper et al., 2022), as they are often universal across cultures. However, cultural differences may still influence the way in which people express or interpret their environmental motives and emotions (Ghali-Zinoubi, 2022), which requires further research.

Technically, our results showed that the subjects displayed a positive attitude towards sustainability when using different implicit measures. This is consistent with other implicit measurement-based studies (e.g., Hietanen et al., 2007; Hietanen & Korpela, 2004; Korpela et al., 2002). Furthermore, our findings indicate that subjects’ associations can be assessed consistently, even if particular measures are not correlated. Previous research has reported similar findings (Bosson et al., 2000; Sánchez et al., 2016).

Previous research has shown that environmental concern seems to be a strong value-based predictor of pro-environmental behaviours and related attitudes in Russia (Sautkina et al., 2021). We assume that this relationship does not involve the affective sphere of implicit attitudes, as we did not find any relation to components of such concern. However, it is possible that implicit environmental attitudes may not be as strong or influential as explicit environmental attitudes and therefore may not have as much of an impact on environmental concern (in line with Brick & Lai, 2018; Sánchez et al., 2016).

Some studies show a compensatory relationship between the implicit and explicit attitudes (e.g. König et al., 2016). Unfortunately, our experimental design does not allow us to test such a hypothesis and we encourage researchers to continue this work. Other potential explanations include the complex nature of explicit high-order attitudes and some difficulties in measuring its latent construct (Sparks et al., 2022), or simply social desirability (e.g., Cameron et al., 2012; Sánchez et al., 2016).

Thus, measuring implicit environmental attitudes can be challenging because they are often unconscious and difficult to detect. People may not be aware of their own implicit attitudes, and even if they are, they may not be willing to share them or do this consistently. Additionally, implicit attitudes are often context-dependent and can change over time, making them difficult to measure accurately. These issues also require further investigation.

We have contributed to the growing evidence that sustainability and pro-environmental behaviour are perceived positively at the individual level, even in the emerging pro-environmental culture of Russia. Using a combination of methods and materials, our study suggests that implicit attitudes associated with pro-environmental behaviour correspond to explicitly positive ones, and there is seemingly no relationship between implicit attitude and explicit environmental concern.

The research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

The author declares that he has no known competing financial interests or personal relationships that could have appeared to influence the work reported in this manuscript.

The author thanks his colleagues and the experts for their help in developing the stimulus materials, valuable advice and discussion of the results: E. Milyaeva, A. Chikvin, O. Golubeva, E. Krapivina, A. Romodina, A. Stetsyuk, A. Bespalov, A. Vlasov, V. Moskvina, M. Gelrud, A. Ivanova; Dr. Cameron Brick for his inspiring research and helpful advice; as well as the editor and independent reviewers for their help and guidance on this work.

All data, materials, and scripts are available at https://osf.io/ayk28/files/osfstorage; https://osf.io/z9d4x.

Environmental culture in Russia is undergoing an active transformation, which has led to some differences in everyday practices. Waste recycling is gradually incorporated in the everyday life of Russians, but at the individual level environmental harm may not yet be clearly understood. Russia lacks both recycling infrastructure and a systematic compulsion to engage people in such behaviour. The purchase of eco-labelled green goods is very common in Russia, as marketers successfully exploit healthy lifestyles and longevity in almost all consumer segments. Russians usually associate the obvious aspects of green consumption with the absence of plastic or other packaging, buying products at food markets or from producers with appropriate eco-labelling (Valko, 2021). Resource-saving behaviour is entirely based on daily routines, but only electricity metres and cold/hot water metres are widespread. The level of pro-environmental transport culture in Russia is much lower (Ratner et al., 2021). Cities have many cars parked everywhere; traffic jams in city centres; insufficient bicycle lanes; special lanes for public transport are usually not maintained, and electric transport has been eliminated in many cities.

Since there is no consistent government policy maintaining pro-environmental behaviour in Russia, it is possible to assume that behavioural costs of such activities are quite high (Ratner et al., 2021). There are a variety of individual pro-environmental practices with sparse involvement that inconsistently shape the landscape of Russian environmental culture. This makes Russia an interesting case to study both from empirical and theoretical points of view. Thus, there is an obvious gap in knowledge of pro-environmental behaviours in Russia and related environmental attitudes. In such a context, further research is needed to understand whether pro-environmental behaviour and how people in Russia perceive it is related to positive attitudes (as it is conventional for Western people (Contzen, Goda, et al., 2021; Contzen, Handreke, et al., 2021; Venhoeven et al., 2020) or simply social desirability.