The literature on environmental management depicts that several theories have been used to understand individual behavioral intentions. The theory of planned behavior (TPB) developed by Ajzen (1991) posits that an individual's intention to engage in a behavior is determined by their attitude toward that behavior, the perceived social norms related to that behavior, and the individual's perceived behavioral control. The norm activation model (NAM) posits that personal and social norms influence an individual's behavior (Onwezen et al., 2013; Goh et al., 2022). Recently, many researchers have used NAM to evaluate individual recycling intentions (Arkorful et al., 2022; Nketiah et al., 2022; Zhang et al., 2022). In addition, the value-belief-norm (VBN) model also explains and predicts recycling intention. According to the VBN model, an individual's behavior is influenced by their values (personal beliefs about what is essential), beliefs, and norms (Stern et al., 1999). VBN posits the belief that recycling effectively reduces waste and preserves resources (Whitmarsh et al., 2018) and the presence of social norms that support recycling (Stern et al., 1999). Previous studies indicate the significance of norms shaping individual pro-environmental behaviors. This study applied Ajzen's (1991) TPB to understand household behavior. The environmental management studies depict that AT, SN, and PBC have a significant role in understanding behavioral intentions (Arvola et al., 2008; Aboelmaged, 2021; Waris et al., 2021). Researchers found that TPB constructs significantly influence individual intention in the context of waste recycling (Khan et al., 2019; Aboelmaged, 2021; Hameed et al., 2022). Therefore, this study extends TPB theory by incorporating IP and MN as additional constructs to evaluate household waste-sorting intention.

Previous studies have found that moral norms significantly predict pro-environmental behavior (Chan and Bishop, 2013; Botetzagias et al., 2015). In the context of pro-environmental behavior, a meta-analysis by Klöckner (2013) showed that MN mediated the relationship between ATD and intention. In relation to this, Liu et al. (2020) conducted a study on Chinese consumer green purchases that revealed MN mediated the relationship between ATD and intention. The mediating effects of MN on the relationships between IP and intention have not been verified in previous studies. Many studies have confirmed the direct impact of MN on recycling intentions (Khan et al., 2019; Soomro et al., 2022). For example, researchers found direct effects of MN on intention (Chen and Tung, 2014). However, studies have not examined the relationship between IP, MN, and intention. Accordingly, this research seeks to find whether moral norms mediate the relationship between information publicity and intention. Hence, we propose the following hypothesis:

Ajzen (1991) indicated that ATD, SN, and PBC are the three main predictors of behavioral intention. Researchers found that ATD, SN, and PBC are significant mediators in the TPB framework (Tarkiainen and Sundqvist, 2005; Kumar et al., 2017; Bourdage et al., 2020). Previous studies indicate that information publicity has a direct and indirect impact via TPB constructs on solid wastes (Soomro et al., 2022), electronic waste recycling intention (Wang et al., 2018), and adoption of solar energy (Waris et al., 2022). Furthermore, studies revealed that MN determines intention and is an important predictor of AT, SN, and PBC (Arvola et al., 2008; López-Mosquera et al., 2014; Soomro et al., 2022). Many recent studies provided empirical evidence of significant indirect effects of MN on intentions via AT, SN, and PBC (Klöckner, 2013; López-Mosquera et al., 2014; Wang et al., 2016; Liu et al., 2020). The literature provides enough evidence on the indirect effect of the MN on behavioral intentions via ATD, SN, and PBC. Thus, it is believed that MN indirectly impacts an individual's WSI. Therefore, the following hypotheses are proposed:

This study proposes a sequential indirect relationship between IP and WSI through MN and TPB constructs (ATD, SN, and PBC) based on the assumption that MN may not directly influence household WSI. It holds that MN significantly influences WSI (Khan et al., 2019; Wang et al., 2020). However, increasing the WSI household must have positive ATD, feel societal pressure, and accept that they can easily perform waste sorting. Wang et al. (2018) findings confirmed that ATD, SN, and PBC are essential mediating factors between IP and intention that signifies the importance of the IP leading to behavioral intention. Other researchers found significant direct and indirect impacts of moral norms and TPB constructs in the context of pro-environmental behavior (Khan et al., 2019; Liu et al., 2020; Soomro et al., 2022). Previous researchers indicated that IP and MN significantly improve behavioral intention. Accordingly, the current study posits that the relationship between IP and household WSI enhances via MN and TPB constructs. Therefore, we propose the following hypotheses:

Figure 1 illustrates the sequential mediating effect of moral norms and TPB constructs (ATD, SN, and PBC) in the relationship between IP and WSI.

This study uses a face-to-face survey technique for the collection of data. Since the present study is focused on WSI by households, therefore purposive sampling technique was employed because it helps in the generalizability of the results where data were collected to ensure a true representation of the population. The main disadvantage of this sampling technique is that bias occurs during the process of sample selection. To minimize bias during data collection, the researcher selected individuals who have an understanding of waste sorting and related concepts. In the past, researchers used purposive sampling techniques to recruit participants who possessed the necessary topic knowledge and understanding of the variables (Khan et al., 2019; Hameed et al., 2022). The questionnaire consists of three parts. The first part was the participants' informed consent stating the study's purpose. This part also ensured the confidentiality of participants' data. The second part was related to demographic information, and the third part collected information on constructs items. We have followed Hair et al.'s (2014) suggested guidelines for calculating the final sample size. The study requires 270 samples (10 responses per item for 27 items). Households were approached in different parts of cities, such as homes, parks, shopping malls, restaurants, and retail stores. Data were collected from 500 individuals from 147 households to enhance our findings' reliability. However, we received many unfilled and multiple response (invalid) questionnaires at the end. The final sample size was 361 after discarding the unfilled and invalid questionnaire. The data collection period ranged from 01 August 2022 to 20 December 2022. Table 1 shows the household demographic information.

This study adopted measurement scales from past studies to collect household data. Three academic experts from the departments of Management, Marketing, and English have evaluated the content of the final questionnaire. A pilot study involving 45 participants was carried out. The pilot study results were satisfactory as all adopted items of the questionnaire were loaded between 0.60 and 0.70, except one item of information publicity construct that was subsequently modified. After receiving recommendations from experts to make the final questionnaire more understandable for the household, the modifications recommended were incorporated. The experts also reviewed and revised the instruments' items to confirm their content validity. They also identified typos and grammatical errors and suggested layout modifications, which were incorporated into the final questionnaire. Table 3 shows the number of items adopted and their sources. The measuring items for the attitude and subjective norms scales were adapted from Tonglet et al.'s (2004) study. The measuring items for the perceived behavioral control scale were adapted from Hameed et al.'s (2022) study. The measuring items for the information publicity scale were adapted from Wang et al.'s (2018) study. The measuring items' MN scale was adapted from Botetzagias et al.'s (2015) study. The measuring items for the waste-sorting intention were adapted from the Echegaray and Hansstein (2017) study. A five-point Likert scale was used to measure all items, where one indicated strongly disagree, five indicated strongly agree, and three indicated a neutral response.

The Kaiser–Meyer–Olkin (KMO) test was employed to assess the sampling adequacy of each variable and the overall model. If the significance of Bartlett's test of sphericity is below 0.5, it indicates that the data exhibit strong construct validity. On the contrary, if the KMO value is ≥0.50, it suggests that the sample is sufficient, and higher KMO values indicate greater adequacy (Qalati et al., 2022a). Table 2 shows sample adequacy results. The KMO values exceed 0.7, and Bartlett's test of sphericity sig value is < 0.05. These results indicate that the sample data are sufficient, demonstrating strong construct validity and making it suitable for analysis.

The presence of common method bias (CMB) poses a risk to the reliability of the gathered data. Therefore, we employed Harman's single-factor test to avoid CMB. The data are not affected by CMB when the amount of variance explained by a single factor is below 50%. The results of Harman's single factor show that the single factor explained a 30.77% variance in the data. Hence, we conclude that CMB is not a threat to the reliability of gathered data (Podsakoff et al., 2003). We also assessed the variance inflation factor (VIF) of all independent variables to ensure that multicollinearity is not a problem. The values of VIF were between 1.390 and 2.926, less than the recommended threshold value of 3.3, depicting no multicollinearity issue (Hair et al., 2021).

The evaluation of the measurement model includes the assessment of reliability and validity. Reliability denotes the consistency and stability of the measuring instrument utilized over time. Validity refers to the extent to which the measuring tool accurately assesses the intended behavior it aims to measure (Yilmaz, 2013). This study uses Cronbach's alpha and composite reliabilities criterion to measure the internal consistency of the data. The values of Cronbach alpha and composite reliabilities above 0.70 indicate internal consistency in the data. Table 3 shows that the study data meet the internal consistency as the values of Cronbach alpha and composite reliabilities were above 0.70. Then, we measured the convergent validity. It shows the degree of constructs relatedness. Convergent validity is confirmed when the average variance extracted (AVE) value is more than 0.50, and factor loading should be at least 0.70 (Hair et al., 2019). The results indicate low factor loadings for the IP3 (0.680) and PBC1 (0.696). Sarstedt et al. (2021) posit that factor loading < 0.70 is common in social science studies. Therefore, we decided not to remove these items. However, the authors suggested the removal of items having low factor loadings (0.40–0.70) from the scale if the deletion of items improves AVE or CR values. First, we followed the same criteria by removing low-loading factors and observed that the removal of a single item resulted in lower factor loadings of the other items such as IP1 and IP2. Thus, following Sarstedt et al. (2021) suggestions, we retained IP3 and PBC1. Table 4 is showing the results of measurement model.

Discriminant validity reflects the unrelated relationship among constructs. In other words, it confirms that all variables are unrelated (Hair et al., 2014). We used two criteria to measure the discriminant validity. Fornell and Larcker's (1981) criterion was the first to assess the discriminant validity of the data. According to this criterion, the value of the square of AVEs should be more than the corresponding correlation coefficients. The results of the discriminant validity as per Fornell and Larcker's (1981) criterion are shown in Table 5. The second criterion was the heterotrait–monotrait ratio (HTMT). According to this criterion, the values of the constructs must be < 0.85 to ensure that all constructs are unrelated to each other (Henseler et al., 2009). Table 6 shows that constructs values are < 0.85 confirming that measured constructs are unrelated.

The quality of the model fit depends on its ability to estimate the endogenous construct (Hair et al., 2014). The main criterion for evaluating the inner model is checking the coefficient of determination (R²) and cross-validated redundancy (Q²; Henseler et al., 2009; Hair et al., 2014). R square (R²) evaluates the accuracy of the model prediction. The variance explained by the independent variables on the dependent variable is represented by R² (Hair et al., 2014). Researchers have categorized R square into high, moderate, and low categories. R² is considered strong if its value is more than 0.6, moderate between 0.3 and 0.6, and low below 0.3 (Sanchez, 2013; Khan et al., 2019). The result indicates the value of R² is 54.6% which is near to considerably high. Cross-validated redundancy (Q²) is another way to assess the model quality. We have also assessed the predictive relevance of the inner model with (Q²) via the blindfolding method. Q² value > 0 confirms the model's predictive relevance (Hair et al., 2014). In this study, the value of Q² is 40.6, which is significantly high, depicting the high quality of the inner model.

To assess the prediction of the PLS model, we performed out-of-the-sample predictive relevance using PLSpredict (Shmueli et al., 2019). We based our predictive evaluation on root mean squared error (RMSE). Next, we conducted a comparison between the items of the endogenous constructs in PLS-RMSE and the items of the endogenous constructs in the linear model RMSE (LM-RMSE). The results show that most of the PLS-RMSE values are lower than the LM-RMSE values, suggesting that the model exhibits a moderate-to-high level of predictive capability for the endogenous constructs (Hair et al., 2019). In addition, PLS-SEM is preferred over LM due to its higher values of Q²_predict for the endogenous items, as shown in Table 7.

The study proposed an extended TPB model to measure household waste-sorting intention. To test the hypothesized model, PLS-SEM was performed using the 2000 bootstrapping sampling technique recommended by the researchers (Haenlein and Kaplan, 2004). The direct and indirect effect results are shown in Tables 7, 8, respectively. The results of the direct relationships show that out of 12 hypotheses, only H1e, which proposed the positive influence of IP on PBC, was not supported (β = −0.005; t = 0.075; p < 0.05). H1a proposed a positive influence of IP on WSI was supported (β = 0.141; t = 3.542; p > 0.05). H1b proposed a positive influence of IP on MN was supported (β = 0.404; t = 7.863; p > 0.05). H1c proposed a positive influence of IP on ATD was supported (β = 0.151; t = 2.542; p > 0.05). H1d proposed a positive influence of IP on SN was supported (β = 0.119; t = 2.057; p > 0.05). H2a proposed a positive influence of MN on WSI was supported (β = 0.106; t = 2.657; p > 0.05). H2b proposed a positive influence of MN on ATD was supported (β = 0.255; t = 4.501; p > 0.05). H2c proposed a positive influence of MN on SN was supported (β = 0.272; t = 4.548; p > 0.05). H2d proposed a positive influence of MN on PBC was supported (β = 0.248; t = 3.866; p > 0.05). H4 proposed a positive influence of ATD on WSI was supported (β = 0.277; t = 5.005; p > 0.05). H5 proposed a positive influence of SN on WSI was supported (β = 0.190; t = 3.427; p > 0.05). H6 proposed a positive influence of PBC on WSI was supported (β = 0.301; t = 6.704; p > 0.05). In addition, the study findings revealed a significant impact of income on WSI (β = 1.48; t = 4.188; p > 0.05). However, the impact of the control variables, age (β = −0.049; t = 1.356; p < 0.05) and gender (β = 0.022; t = 0.626; p < 0.05), on WSI was insignificant. Overall, the extended TPB model provides a comprehensive framework for understanding household WSI. Figure 2 is showing the results of structural equation model. These findings show that IP is important in creating awareness and establishing moral norms regarding the proper disposal of waste. With the help of education programs and campaigns, households can learn about waste-sorting benefits such as conserving natural resources, reducing the amount of waste, and preserving the natural environment.

The indirect effects were estimated using the bootstrapping technique to examine the mediating role of MN and TPB constructs (ATD, SN, and PBC). We utilized the variance accounted for VAF formula to assess the indirect impact of IP and MN, a mediation analysis approach suggested by Hair et al. (2016). Recently, several studies have used the VAF approach to test mediating effects (Nitzl et al., 2016; Rasoolimanesh et al., 2021; Qalati et al., 2022b). The formula of VAF is given as follows:

where the total effect is the sum of the direct and indirect effects. According to Hair et al. (2016), VAF values below 20% indicate no mediation, values between 20 and 80% indicate partial mediation, and values above 80% indicate full mediation. The following are the VAF values for the present study:

The results revealed that 23.3, 39.7, 32.9, 41.4, 20.1, and 24.5% indirect effects of H3, H7a, H7b, H7c, H8a, and H8c, respectively, indicate partial mediation, whereas H8b result shows 12.9% indirect effect, indicating no mediation.

This study contributes to the TPB theory and enriches the environmental management literature. While earlier research has applied the extended TPB framework to predict waste management, waste separation, and recycling intention (Khan et al., 2019; Hameed et al., 2022; Soomro et al., 2022; Zhang et al., 2022), this study presents an extended TPB framework that simultaneously assesses the impact of IP and MN on waste-sorting intentions. Zhang et al. (2015) and Wang et al. (2018) posited that IP significantly influenced residents' recycling and waste separation intentions. Therefore, they call for exploring relationships between IP and other antecedents of waste management. In addition, prior studies have only assessed the direct and mediating impacts of studied constructs. The current study has proposed a novel framework that assessed the mediating impact of MN. In addition, the current study has proposed serial mediation analysis and enriched the understanding of the relationships among IP, MN, and TPB constructs. The addition of IP and MN is a vital contribution that significantly enhanced the variance in the extended TPB model.

Practically, this study presents several implications to managers and practitioners who contribute to the betterment of the environment. First, this study reveals that PBC is the most influential factor in shaping waste-sorting intention among households. PBC relates to an individual's ability and tendency to perform a given task. From the study's outcomes, it is clear that households think they can contribute to environmental sustainability through waste-sorting efforts. Therefore, managers and practitioners should focus on the arrangements of campaigns that transform beliefs into behavior and provide helpful information related to waste sorting. The results also depict that MN is an important antecedent of waste-sorting intentions, signifying the importance of household acceptance of responsibility toward protecting the environment. Due to the mismanagement of household waste, the challenge of managing increasing waste is becoming more problematic for the governments of developing nations (Khan et al., 2019; Hameed et al., 2022). Therefore, it is recommended that collective and concentrated efforts be made to promote waste-sorting activities. Researchers argued that people in developing countries have comparatively lower education levels than people in advanced countries (Khan et al., 2019). Therefore, they pay less attention to environmental issues that depict the mismanagement of waste endangering environmental sustainability. Therefore, it is recommended that companies' products for household use should mention the methods of discarding that will eventually lead to environmental protection. The findings indicate the significant impact of income on WSI and show that high-class households are more conscious of waste-sorting activities. Therefore, policymakers should emphasize the middle-class and low-class segment waste-sorting activities to promote environmental sustainability because most of the developing country population consists of middle- and low-class households. To promote waste-sorting activities, the government can incentivize households and encourage them toward waste sorting. The incentive may include rebates on electricity and water sewerage and discounts on garbage collection fees, for households that properly sort their waste. In addition, the government can collaborate with local communities and companies to provide more education and awareness and help create more accessible and convenient ways for households to sort and manage their waste.