國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
19
Empirical Model and Results
The analysis uses the following model to examine the size of wage gap between the two consecutive years for each group after the employment form changes. The primary empirical model is:
𝑌𝑖 = 𝛼 + 𝛽1𝑆1𝑖+ 𝛽2𝑆2𝑖+ 𝛽3𝑆3𝑖+ 𝑋𝑖′𝜆 + 𝜀𝑖 , 𝑖 = 1, … , 𝑁 (1)
Where 𝑌𝑖 is the difference between wages in the two consecutive years for respondent i, that is, the monthly income in the second year minus the monthly income in the first year for each individual. The variables, 𝑆𝑖, are a set of dummies indicating the group each individual i belongs:
𝑆1𝑖: This variable indicates workers employed in temporary work for two consecutive years, worker i.
𝑆2𝑖: This variable indicates workers employed in temporary work in the first year but switching to non-temporary work in the second year, worker i.
𝑆3𝑖: This variable indicates workers employed in non-temporary work in the first year but switching to temporary work in the second year, worker i.
And 𝛽1, 𝛽2 and 𝛽3 are the impacts of these groups by job status. One job status dummy is dropped from the equation, the group where workers were employed in non-temporary work for two consecutive years.
The vector 𝑋𝑖 contains the characteristics of individual i: sex, age and its square, marital status, educational level, geographic location, years, and the industry of the job, and all these are used to control for the individuals’ characteristics.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
20
Table 8 presents OLS estimates for difference between wages in two consecutive years and the groups by job status and the characteristics of samples. Column (I) reports the estimation without controls. For workers in group 𝑆2, who had temporary work in the first year and switched to non-temporary work in the second year, change of job status seems to bring benefits to monthly income. The average monthly income for workers in group 𝑆2 is
$1,194 NTD, about 4.98% of average monthly income, larger than workers employed in non-temporary jobs for two consecutive years.
Workers in group 𝑆3 taking up non-temporary work in the first year and switching to temporary work in the second year earned less after the change of job status. The estimate shows that the average monthly income in group 𝑆3 is $2,234 NTD, about 5.9% of the monthly income, less than workers taking up non-temporary jobs for two consecutive years.
Column (II) in Table 8 shows estimation with control variables. The estimates in the two columns do not considerably differ from each other. Compared to workers employed in non-temporary jobs for two consecutive years, the average monthly income for workers in group 𝑆2 is larger by $1,296 NTD, and the average monthly income in group 𝑆3 is lower by $2,197NTD. The changes take up about 5.4% of the average monthly income for group 𝑆2 and about 5.8% for group 𝑆3.
Using the OLS estimates to examine the relationship between the change of job status and monthly income, the results show that change of the form of employment may hurt wages, especially after switching job from non-temporary to temporary while change from temporary to non-temporary benefits the workers.
‧
The Wage difference and changing in job status (I) OLS estimation
Kleibergen-Paaprk Wald F statistic 9.278
N 45844 45844 45844
NOTE: The OLS estimation using the robust option, that is, the standard error concerns the heterogeneity and lack of normality. The instrumental variables estimation using heteroskedasticity-based instruments and the robust option.
∗∗∗Significant at the 1 percent level.
∗∗Significant at the 5 percent level.
∗Significant at the 10 percent level.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
22
However, when there exists endogeniety in the model, the OLS estimation can become biased and inconsistent. The non-zero correlation between the disturbance 𝜀 and job status causes endogeneity. Some variables correlated with the independent variable in the model are omitted because these variables are unobserved, for example, work attitude.
Work attitude influences both workers’ decision about choosing job status and their monthly wages. Proactive workers usually prefer stable jobs and they expect employee training, struggle for promotions, have higher job satisfaction and care about work circumstances. Thus, non-temporary jobs are more suitable for them. This kind of workers take up non-temporary jobs or strive for switching to non-temporary jobs. Also, workers who are proactive apply themselves to their tasks and this may make them get higher paid.
On the other hand, preferences of passive workers in terms of stable jobs, opportunities of promotion or any other work condition are not quite clear. So it is more likely that this kind of workers accept temporary jobs. The passive work attitude also induces lower monthly incomes since they pay less attention to their work.
In this case, work attitude is a determinant of employment form and the wage level, that is, it simultaneously affects the dependent and independent variables in the model and it should not be left out. However, work attitude is a subjective cognition which is not easy to determine and there is no information about it in the database. Therefore, work attitude is an unobserved variable and causes omitted variable bias.
When there exists endogeneity in the model, the instrumental variables can be used to eliminate the bias. To serve as valid instruments for the endogenous variables, instrumental variables must satisfy two conditions. First, the instruments must be exogenous, that is, the
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
23
covariance between the instrument and the disturbance 𝜀 should be zero. In addition, the instrument must be correlated with the endogenous variable, which means that the covariance between the instrument and the endogenous variables should not be zero.
In order to deal with the endogeneity in the model, finding a valid instrument is a good way. The model for adding instrumental variables can be rewritten as:
Y𝑖 = 𝛼1𝑖+ 𝛽1𝑆1𝑖+ 𝛽2𝑆2𝑖+ 𝛽3𝑆3𝑖+ 𝑋𝑖′𝜆 + 𝜀𝑖 , 𝑖 = 1, … , 𝑁 (2) 𝑆1𝑖= 𝛼2𝑖+ 𝑧1𝑖′𝛾1+ 𝜀1𝑖
𝑆2𝑖 = 𝛼3𝑖+ 𝑧2𝑖′𝛾2+ 𝜀2𝑖 𝑆3𝑖 = 𝛼4𝑖+ 𝑧3𝑖′𝛾3+ 𝜀3𝑖
where 𝑧1𝑖, 𝑧2𝑖, and 𝑧1𝑖 are the instruments and 𝜀𝑖, 𝜀1𝑖, 𝜀2𝑖, and 𝜀3𝑖 are the disturbances in the model.
In this study, the suitable instrument needs to be uncorrelated with the disturbance 𝜀𝑖 and be correlated with the job status. However, no appropriate instruments could be found in the data set. Thus, in this study, the method provided by Lewbel (2012) is used to create instruments.
Lewbel assumes that the
E(𝑋𝑖𝜀𝑖) = 0,
E(𝑋𝑖𝜀𝑗𝑖) = 0, 𝑗 = 1,2,3,4; 𝑖 = 1, … , 𝐼 Cov(𝑧1𝑖, 𝜀𝑖𝜀1𝑖) = 0, Cov(𝑧2𝑖, 𝜀𝑖𝜀2𝑖) = 0,
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
24
Cov(𝑧3𝑖, 𝜀𝑖𝜀3𝑖) = 0, i = 1, … , I
And 𝜀𝑖 and its variance-covariance matrix have heteroskedasticity.
In the study, there are three endogenous regressors, 𝑆1𝑖, 𝑆2𝑖 and 𝑆3𝑖, but I cannot find any available instruments. Under the assumptions mentioned above, Lewbel (2012) indicates that the instruments can be created from the residuals of the first-stage regression and included as exogenous variables. The constructed instruments are as shown below:
𝑧1𝑖 = (𝑋𝑖 − 𝑋̅)𝑒1, 𝑧2𝑖 = (𝑋𝑖 − 𝑋̅)𝑒2, 𝑧3𝑖 = (𝑋𝑖 − 𝑋̅)𝑒3.
where 𝑒1, 𝑒2, and 𝑒3 are the residuals from the first-stage regression of the endogenous regressors 𝑆1𝑖, 𝑆2𝑖 and 𝑆3𝑖 on all exogenous regressors.
Column (III) of Table 8 shows instrumental variables estimates of Equation (2) for the impact of job status on income differences. The estimates reveal that the average monthly income of workers employed in temporary jobs for two consecutive years are $718.6 NTD less than workers who were employed in non-temporary jobs for two consecutive years, about 3% of the average monthly income. Average monthly income of workers who had non-temporary jobs in the first year and changed job status to temporary jobs in the second year is $2,476 NTD less than workers who had non-temporary jobs for two consecutive years, about 6.5% of average monthly income, workers who had temporary jobs in the first year and changed job status to non-temporary jobs in the second year is $1,510 NTD more than workers who had non-temporary jobs for two consecutive years, about 6.2% of average monthly income.
‧
The Hansen J statistic for testing the over-identification for all instruments is 117.47, and the p-value is 0.0093. At 95% confidence level, the null hypothesis that the model is valid is rejected, which means the instruments set is not that appropriate. Weak identification test is about instrument relevance and an instrument is considered weak when the correlation between the instrument and the endogenous variable is weak. The Kleibergen-Paaprk Wald F-statistic is used for checking whether weak identification exists and the rule of thumb indicates that the F-statistic should be greater than 10 or we have to worry about weak instruments. The Kleibergen-Paaprk Wald F-statistic in the model is 9.057, and according to the rule of thumb mentioned above, weak identification may occur. Weak instruments bring about biased results, so the IV estimation shows incorrect coefficients and significances.
Table 9 presents the estimated impact of job status change on monthly income difference for different age groups. For workers in group 𝑆3, there are statistically significant at an error level of 5 percentage or less for all age groups except over sixty years of age. Workers who are under forty years of age suffer lower damage than those who are over forty years of age, and the differences between workers over and below forty years of age are higher than
$1,000 NTD.
Younger workers have less work experience and usually receive lower wages than older ones, so when they change their job status from non-temporary to temporary, the wage difference is smaller than older workers. Workers who are beyond forty years of age suffer about $3,000 NTD wage loss when they change their jobs forms from non-temporary to temporary, which is apparently greater than workers less than forty years of age suffer.
Almost all workers significantly suffer from wage loss after change from non-temporary to non-temporary job and the income loss gets worse as the age increases, as shown
‧
The monthly income difference and change in job status by age IV estimation with controls
Location included included Included included included
Industry included included Included included included
Years included included Included included included
constant 532.9 19,916 -11,103 -41,447 -56,313
(4044.00) (14906.00) (58451.00) (82970.00) (78338.00)
Hansen J statistic 88.53 90.65 102.93 81.60 74.78
Kleibergen-Paaprk WaldF statistic
5.82 6.04 10.30 3.87 20.84
N 9035 13016 12860 8551 2382
NOTE: The instrumental variables estimation using heteroskedasticity-based instruments and the robust option. The robust option means that the standard error concerns the heterogeneity and lack of normality.
∗∗∗Significant at the 1 percent level.
∗∗Significant at the 5 percent level.
∗Significant at the 10 percent level.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
27
in Table 9, though there is no significant change for workers over sixty years of age. One possible explanation for this result may be that there are few observations for this group which leads to no significance. The other explanation is about their work experiences and bargaining power. Workers beyond sixty years of age have rich work experience so employers would not take them as less productive workers. Also, these workers usually employed in temporary jobs voluntarily, thus they still have bargaining power on their wage.
Workers under forty years of age receive significant wage increase after moving from temporary to non-temporary jobs. Young workers are more amenable to take up temporary jobs and they often change their jobs. Because of lack of work experience and stability, young workers receive lower wages. So when they switch to non-temporary jobs, they can significantly benefit from it. But the amount of wage increase for workers between thirty and forty years of age is smaller than workers under thirty years of age, implying that this impact decreases as age increases (Table 9). There is no evidence of workers over forty years old gaining from job status changing from temporary to non-temporary. Table 10 shows the OLS estimation of the monthly income difference by change in job status by age. The outcomes are basically same as the IV estimation.
The p-values of Hansen J statistic are not significant for any of the age groups, suggesting that the null hypothesis of over-identification for all instruments cannot be rejected and this model is valid. The Kleibergen-Paaprk Wald F-statistic needs to be larger than ten or there would be no evidence to state that weak instruments do not exist. In Table 9, only the over 60 years age group has an adequately large F-statistic to reject the null hypothesis; in all other groups weak instruments may exist and lead to biased results.
‧
The monthly income difference and change in job status by age OLS estimation with controls
Location included included Included included included
Industry included included Included included included
Years included included Included included included
constant 1378 22010 -12418 -39732 -60,024.3
(4068) (14919) (58372) (83103) (78419.89)
R-squared 0.008 0.005 0.003 0.005 0.013
N 9035 13016 12860 8551 2382
NOTE: The OLS estimation using robust option, which means that the standard error concerns the heterogeneity and lack of normality.
∗∗∗Significant at the 1 percent level.
∗∗Significant at the 5 percent level.
∗Significant at the 10 percent level.
‧
The monthly income difference and change in job status by gender
OLS estimation IV estimation
(I)Female (II)Male (III)Female (IV)Male
(𝑆1) Temp to Temp -384.9 -301.5 -677.6 -626.4
Location included Included included included
Industry included Included included included
Years included Included included included
constant 1,015 340.9 -546 -1,361
(1268.00) (2124.00) (1314.00) (2119.00)
R-squared 0.003 0.004 . .
NOTE: The instrumental variables estimation using heteroskedasticity-based instruments and the robust option.
The robust option means that the standard error concerns the heterogeneity and lack of normality.
∗∗∗Significant at the 1 percent level.
∗∗Significant at the 5 percent level.
∗Significant at the 10 percent level.
‧
Table 11 reports estimated coefficients of the impact of job status changing on monthly income for male and female workers and the results show that the impacts are different for the two genders. The IV estimates show that for male workers, changing from non-temporary to temporary results in significant damage to the monthly wage. On average, they suffer about $3,519 NTD wage loss. This loss accounts for 8.47% of monthly income, which is really a considerable damage to male workers. While for female workers, there is no evidence showing that changing job status from non-temporary to temporary does significant harm to their monthly income. There is about $2,061 NTD average increase in monthly income, which is about 10.5% of monthly income, when female workers change from temporary to non-temporary but no evidence shows the same result for male workers.
The influence of job status change differs across genders mainly due to the standard roles in economics of the family. In most cases, wife is the one staying home and taking care of children. Thus, women leave labor force or take up temporary jobs. Women take temporary jobs because of flexibility, which allows them to work at home or the time for work does not have to be a period of long and continuous hours. So women still have the bargaining power and employers do not take them as less productive workers who cannot find better jobs. Therefore, the wage loss for women on change of job status from non-temporary to non-temporary does not show a significant decrease.
When female workers switch from temporary to non-temporary jobs, their wage can significantly increase. Becoming non-temporary workers means that these female workers’
productivity is approved and because of the family role, females take temporary jobs do not lead to severe discrimination thus employers are willing to offer these female workers higher wage.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
31
In contrast, male workers changing job status from non-temporary to temporary usually do not do so voluntarily and this makes the circumstance different from female workers.
Employers see these male workers as less skilled or having lower productivity and therefore are not willing to offer good pay. As a result, male workers suffer severe wage loss when switching from non-temporary jobs to temporary jobs.
When male workers move from temporary to non-temporary jobs, employers see these workers previously employed in temporary job as a sign of low productivity, so even when they are employed in a non-temporary form, there is no significant benefit in terms of wages.
That is, the wage penalty for males having taken up temporary jobs previously is more serious than females.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
32
Conclusions
The increase in number of temporary workers in Taiwan has led to concerns about its influence on workers. Employment of temporary form really increases labor market flexibility and brings about benefits for employers and some workers. Employers can use temporary workers to meet temporary needs to reduce cost. Some workers take up temporary jobs because they need short-term contract or some other kind of flexibility that only temporary jobs can provide. But this kind of employment form leads to damage to workers.
Workers in temporary jobs usually have lower wages, less chance for promotion, lower job satisfaction and this may influence the wages in subsequent jobs also.
Using the OLS estimation and Lewbel’s (2012) method, which is used to create instruments to resolve endogeneity, this paper investigates the impact of change in job status on monthly income in Taiwan. Different from previous empirical studies, workers are divided into only two forms of employment, temporary and non-temporary jobs. For comparing the differences between them, this study divides all workers into four groups according to their job status in two consecutive years and examines the size of income change caused by the change in job status.
The results suggest that on average, compared to workers having had non-temporary jobs for two consecutive years, workers who move from non-temporary to temporary jobs suffer significant income losses, $2,197 NTD, about 5.8% of the average monthly income.
Workers who move from temporary to non-temporary jobs receive significant additional income of $1,296 NTD a month, about 5.4% of the average income.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
33
Moreover, the influence of job status changing varies across gender. Female workers do not suffer significant income loss when changing job status from non-temporary to temporary, and can receive significant benefits when they switch from temporary to non-temporary jobs. The traditional family role makes them give up their current jobs for staying at home or taking up temporary jobs to take care of the family, so they take up temporary jobs voluntarily and have bargaining power on wages.
For male workers, changing from non-temporary to temporary does significant damage to the monthly wage, but there is no significant benefit from switching from temporary to non-temporary jobs. Men are less inclined to take up temporary jobs since they usually are the breadwinners in the family. So when they accept temporary jobs it means they have few alternatives, implying low bargaining power. In addition, having taken a temporary job previously is a signal of less productive workers.
Workers’ between 40 and 60 years of age suffer more severe wage loss than those under forty years when changing from non-temporary to temporary jobs. Workers in this age group reach the top of the wage level and have much more influence on their wages when switching job status from non-temporary to temporary. Young workers receive significant income increase when they change job status from temporary to non-temporary. Lack of experience and stability result in young workers not getting good pay. Once they become non-temporary workers, they obtain the recognition from employers and get higher income.
Workers suffer income loss due to lack of productivity and receive less or even no job training during temporary jobs, which makes temporary workers become further disadvantaged. Government can provide some professional training or some subsidies for