The population for this study consists of approximately 5,200 frontline staff employed at twenty five (25) ministries located in the Federal Territory of Kuala Lumpur and Putrajaya, Malaysia. The ministries in Kuala Lumpur and Putrajaya are selected because all the 25 ministries under Malaysian government have their main ministries’ office at either Kuala Lumpur or Putrajaya. The frontline staffs are chosen because they are the government's main representatives who interact directly with customers or the public. Therefore, they have heavy responsibilities and play very important roles in public perception of the public sector or government, in general. Since one of the researchers of this study is a government employee working in one of the ministries, access to sample elements (frontline staff) has not been an issue. The questionnaire was distributed to 625 frontline staff working at the service counters in 25 ministries and the respondents were selected randomly. These counter staff act as a bridge between the citizens and the ministries. Out of 625 questionnaires distributed, 519 (response rate=83%) were returned and 512 were deemed useful for further analysis and these had no missing values. The original English version of the questionnaire was translated into Bahasa Malaysia (local language) using the back translation method with the help of a panel of language experts from the language faculty at a local public university.

EI. The items under this construct were adopted from the scale developed by Wong and Law (2002). This construct had four dimensions: self-emotion appraisal, other's emotion appraisal, use of emotion and regulation of emotion. The scale consisted of 16 items in a five-point Likert format where one corresponded to “strongly disagree” and five corresponded to “strongly agree”. High average scores correspond to high levels of EI.

Personality traits. The items under this construct were adopted from the scale developed by Rammstedt and John (2007). This construct had five dimensions: extraversion, agreeableness, conscientiousness, neuroticism, and openness. The scale consisted of 10 items in a five-point Likert format where 1 corresponded to strongly disagree and 5 corresponded to strongly agree.

Affectivity. The items under this construct were adopted from the scale developed by Watson, Clark, and Tellegen (1988). This construct had two dimensions: positive and negative affectivity. The scale consisted of 20 items in a five-point Likert format where 1 corresponded to not at all and 5 corresponded to extremely. High average scores correspond to high levels of affectivity.

Emotional Labor. The items under this construct were adopted from the scale developed by Grandey (2003). This construct had two dimensions: surface acting and deep acting. The scale consisted of eight items (five items under surface acting and three items under deep acting). High average scores correspond to high levels of emotional labor.

Emotional Exhaustion. The items under this construct were adopted from the scale developed by Maslach and Jackson (1986). The scale consisted of nine items. The scale employed a seven-point Likert scale that ranged from never to everyday. High mean scores on this measure suggest high levels of emotional exhaustion.

CWB. The items under this construct were adopted from the scale developed by Spector et al. (2006). This scale had two dimensions: CWB-O (organizational) and CWB-I (interpersonal). The scale consisted of 45 items. The scale employed a five-point Likert scale that ranged from never to everyday. High average scores correspond to high levels of CWB. The final questionnaire and the data can be obtained from the corresponding author.

A pilot test was carried out to determine the clarity of the instructions and items in the questionnaire. According to Sekaran and Bougie (2010), a reliable instrument should avoid long complex questions, double barreled questions, culture jargon, leading questions, double meaning, and sensitive questions. The researcher conducted a pilot test with 30 participants from one selected ministry. No changes were made to the questionnaire after the pilot test. The questionnaire contained questions related to frontline staffs’ EI, personality, affectivity, emotional labor, emotional exhaustion, and CWB. Before the questionnaires were distributed to the frontline staffs, a formal approval was obtained from the related ministry. Later, the questionnaires were given to the officer in-charge to distribute to the selected frontline staffs that were identified randomly. The officer in-charge was chosen to expedite the process of distribution and collection of questionnaires. The reliability scores of the constructs ranged between .70 and .92.

The reliability of the constructs was measured using Cronbach alpha. The reliability scores are given in Table 1 and it can be seen that the scores of the constructs are between .70 and .98. The validity of the constructs was tested based on the results from Confirmatory Factor Analysis (CFA), which was performed using Lisrel 9.1 student version. We used the following criteria for the convergent and divergent validity tests (Hair, Black, Babin, & Anderson, 2010; Podsakoff, Mackenzie, Lee, & Podsakoff, 2003): (1) factor loading of items on the latent construct must be greater than .5, (2) composite reliability (CR) of each construct must be greater than .7, (3) average variance extracted (AVE) of each construct must be greater than .5 and (4) AVE of each construct must be greater than the squared correlation of that construct with other constructs. The validity test results, CR, and AVE are given in Table 1. The AVE scores for personality, emotional labor, and affectivity are below the cutoff value of .5. However, these values are above the squared correlation with other constructs.

In this research, the responses to the questionnaire items have been obtained from a single source and both the dependent and focal explanatory variables are perceptual measures derived from the same respondent. This may result in common method variance. Common method variance is the “variance that is attributable to the measurement method rather than to the constructs the measures represent” (Podsakoff et al., 2003, p. 879). We performed a one-factor Herman test by loading all the items on to a single factor. Factor analysis indicated that this single factor could explain only 29% of the total variance. Podsakoff et al. (2003) indicate that there is no agreed cut-off point but a value of less than 50% is considered to be a reasonable value to indicate the reduced effect of common method variance. Since in this research only 29% of the total variance is explained by a common factor, we can argue that effect of common method variance is limited.

Among the 512 respondents, 70% were females. This is not surprising given the fact that more than 60% of the university graduates in Malaysia are females. According to World Bank report on Malaysia, 68.3% of females are employed in services. About 85% of the respondents are under 35 years of age, 90% of the respondents have less than 10 years of experience and 89% of the respondents have been as frontline employees for less than six years. The mean and standard deviation of all constructs for male and female employees are given in Table 1. Based on the descriptive statistics the following conclusions can be drawn: (1) the mean level of EI can be considered to be ‘high’ (male: mean=4.74, SD=0.83; female: mean=4.72, SD=0.77), (2) the mean extraversion score can be considered to be ‘moderate’ for male and ‘high’ for female (male: mean=3.91, SD=0.75; female: mean=4.15, SD=0.78), (3) the mean agreeableness score can be considered to be ‘moderate’ for male and ‘high’ for female’ (male: mean=3.83, SD=0.85; female: mean=4.04, SD=0.77), (4) the mean conscientiousness score can be considered to be ‘moderate’ for male and ‘high’ for female (male: mean=3.80, SD=0.80; female: mean=4.09, SD=0.66), (5) the mean neuroticism score can be considered to be ‘low’ for both male and female (male: mean=2.39, SD=0.77; female: mean=2.28, SD=0.80), (6) the mean openness score can be considered to be ‘moderate’ for male and female (male: mean=3.06, SD=0.62; female: mean=3.06, SD=0.61), (7) the level of positive affectivity is ‘moderate’ (male: mean=3.73, SD=0.62; female: mean=3.76, SD=0.66) and the level of negative affectivity is ‘low’ (male: mean=2.57, SD=0.67; female: mean=2.51, SD=0.65), (8) the level of emotional labor (surface acting) is ‘moderate’ (male: mean=3.75, SD=0.89; female: mean=3.94, SD=0.81), (9) the level of emotional labor (deep acting) is ‘moderate’ (male: mean=3.79, SD=1.02; female: mean=4.00, SD=0.84), (10) the level of emotional exhaustion can be considered to be ‘low’ (male: mean=1.63, SD=1.29; female: mean=1.60, SD=1.25), (11) the level of CWB can be considered to be ‘low’ (male: mean=1.36, SD=0.55; female: mean=1.25, SD=0.46), and (12) gender differences are significant for personality, emotional labor (surface and deep acting) and CWB; personality trait score and emotional labor are higher for females but CWB is lower.

We have managed to capture 45 different CWB activities indulged by the respondents in the last year. Table 2 gives the exhaustive list of activities and the frequency of indulgence. The frequency indicates that the problem of CWB among front-line government employees may need attention. In order to understand the differences in personality traits and CWB among the respondents in various ministries, we have computed the average levels of five dimensions of personality traits and CWB. Table 3 gives the average levels. The levels indicate that (1) levels of extraversion, agreeableness, and conscientiousness can be considered to be ‘moderate’ to ‘high’, (2) levels of neuroticism and openness can be considered to be ‘low’ to ‘moderate’ and (3) level of CWB can be considered to be ‘low’.

The correlations between the dimensions of all constructs for male and female are given in Table 4. Based on the correlations, it can be observed that dimensions of CWB have significant correlations with dimensions of other constructs. Similarly, dimensions of emotional labor and emotional exhaustion have significant correlations with the dimensions of personality and EI.

The hypotheses were tested using Lisrel 9.1 version. The model fit statistics are: root mean square error approximation (RMSEA)=.042 (acceptable level<.08); chi-square/degrees of freedom=1.826 (acceptable level<3); p-value=.895 (acceptable level<.05); normed fit index (NFI)=.968 (acceptable level > .9); comparative fit index (CFI)=.985 (acceptable level > .9); root mean residual (RMR)=.030 (acceptable level<.08); goodness-of-fit index=.977 (acceptable level > .9). The significant results of the hypotheses testing are given in Table 5. We highlight some interesting results that were contrary to our hypotheses.

First, the negative relationship between positive affectivity and CWB of frontline employees is not supported (β=.049, p-value=.286). Counter to results by Miles et al. (2002) and Dalal (2005), our study shows a positive relationship between positive affectivity and CWB. Second, the hypothesis that establishes the relationship between emotional labor (surface acting) and CWB is not supported (β=.061, p-value=.144). Our result is contrary to the result by Bechtoldt, Welk, Zapf, & Hartig (2007). According to them, CWB is used as the avenue by employees to escape emotional dissonance and this is central to emotional labor (surface acting). Third, the hypothesis that establishes the indirect link between EI and emotional exhaustion through emotional labor (deep acting) is not supported (β=.072, p-value=.095). Our finding contradicts the finding of Cote (2005) and according to him the individuals with high EI are more likely to engage in deep acting during interpersonal interactions. Fourth, the indirect negative relationship between EI and CWB through emotional labor (surface acting) is not supported. In fact, our test reveals that there is a strong positive relationship between EI and emotional labor (surface acting) (β=.367, p-value=.000). Our result is partially supported by Wong and Law (2002). They have argued that employees high on the appraisal of self-emotions and other's emotions recognize the need to engage in emotional labor in a positive manner. But, the authors have not looked at the individual dimensions of emotional labor. Fifth, the indirect negative relationship between EI and CWB through emotional exhaustion is not supported (β=-.058, p-value=.209). Our result does not support the findings by many researchers (For example, Brackett et al., 2010). Sixth, the relationship between emotional exhaustion and emotional labor (deep acting) is not supported (β=-.070, p-value=.114). However, there is a negative relationship between emotional exhaustion and emotional labor (surface acting) (β=-.212, p-value=.000) contrary to our assertion that there is a positive relationship. A few researchers have shown a positive relationship between emotional exhaustion and emotional labor (surface acting) (Chooi, 2012; Johnson & Spector, 2007; Lee, 2010). However, Chu (2002) and Pugh, Groth and Hennig-Thurau (2011), have argued a negative relationship between emotional labor (surface acting) and emotional exhaustion. Seventh, the hypotheses that establish the links between affectivity (positive and negative) and emotional labor (surface acting and deep acting) are not supported. Kammeyer-Mueller et al. (2013) have shown a moderate relationship between these two constructs. Eighth, the hypotheses that test the relationships between positive affectivity (β=.108, p-value=.007), negative affectivity (β=.095, p-value=.021) and emotional exhaustion are partially supported. We have hypothesized the relationship between positive affectivity and emotional exhaustion to be negative. The indirect role of emotional labor between affectivity and emotional exhaustion is not supported since the relationship between affectivity (positive and negative) and emotional labor is insignificant.

Since the percentage of women is higher in our sample (70%), we have performed independent t-test as an additional analysis to study the effect of gender on all the constructs. Based on our results, there are significant differences and these are: (1) extraversion (t-value=-2.789, p-value=.005) with women recording a higher level than men, (2) agreeableness (t-value=-2.731, p-value=.007) with women recording higher level than men, (3) conscientiousness (t-value=-3.899, p-value=.000) with women showing higher level than men, (4) levels of emotional labor (surface acting) (t-value=-2.366, p-value=.018) with women recording a higher level than men, (5) levels of emotional labor (deep acting) (t-value=-2.446, p-value=.015) with women recording a higher level than men, and (4) levels of CWB (t-value=2.328, p-value=.021) with men recoding a higher level than women.

Since the frontline employees are “barometers of the business” (Dagger et al., 2013, p. 498), the managers must ensure that they possess or are trained to have the right skills to perform their jobs (Dagger et al., 2013; Porath, MacInnes, & Folkes, 2011). The selection of frontline employees who are constantly in touch with the common citizens must be done with utmost care since the effectiveness of service delivery depends upon them. The citizens form opinions about the government institutions based on their experiences with the frontline employees. Therefore, it is imperative that these employees with the following characteristics be allowed to handle the citizens: (1) high on personality traits such as extraversion, agreeableness, openness, and conscientiousness and low on neuroticism; (2) high levels of EI, and (3) low levels of negative affectivity. These characteristics will help in managing the emotional labor, emotional exhaustion, and thereby CWB of the employees.

As suggested by Grandey (2000), environment in the form of organizational support plays an important role in managing emotions and stress of service employees. How can the managers help in avoiding unpleasant incidences that may occur when the frontline employees and common citizens are interacting? We outline some of the steps that can be carried out by the managers handling frontline employees.

First, a proper training on enhancing EI of employees needs to be provided. According to Beigi and Shirmohammadi (2011) and Cherniss and Adler (2000), training given to employees helps them handle their emotions better. Second, managers can give job autonomy and therefore, emotional autonomy to employees. Some studies have shown that job autonomy minimizes the stress due to emotion regulation process (Grandey, 2000). Third, managers must support their frontline employees. An employee's perception that he/she is supported by managers and co-workers can help him/her have positive views about the working environment and this in turn can reduce emotional labor and emotional exhaustion (Grandey, 2000; Howes, Cropanzano, Grandey, & Mohler 1999). Fourth, emotional displays (emotional labor) adequate for a particular situation are dependent on culture and situation, and hence learnable (Lazanyi, 2011). A social environment at work that is conducive can help employees identify themselves with the organization and its goals and this will lead to internalization of organizational norms relating to emotional displays in place. Fifth, when the number of women employees is significant as in the case of Malaysia, organizations can provide additional support such as a child care center in the office premises and this may reduce work-family conflict among women employees and this in turn can help reduce their emotional exhaustion and instances of CWBs.