Влияние экономики краткосрочных контрактов на зависимый труд Великобритании

Thirdly, it is important to note that the variable related to the percentage of recent internet users in the last three months has a significant positive correlation with the appearance of the gig economy participant. The 1 percent increase in the number of internet users increases the number of gig economy participants by 0,88 percent. This is quite an intuitive relationship because when more people are using the internet, the possibility that they would enter the form of labor that is connected to the digital platforms is higher.

From the regression model, it could be seen that other variables, such as the job density, the median age in the region, do not have a significant correlation with the development of the gig economy. Constant in this model does not have an important meaning and a direct interpretation.

> Second stage

In the second stage of the two-stage least squares model, the dependent variable (the number of employees) is regressed on the predicted value of the endogenous variable, on the exogenous and control variables. The results of the regression are depicted in Table 5, the regression (1).

***Insert Table 5 here***

First of all, the correlation between the gig economy and the dependent labor should be interpreted, as it is the main question under the investigation in the paper. It could be seen that there is a significant positive correlation between the number of gig firms and dependent employment. The relationship is significant even under the 1% significance level. The relationship shows that when the number of gig firms in the economy increases by 1 percent, the number of employees in the economy increases by 0,0576 percent, implying the weak, but positive relationship between two variables. The effect could be considered to be very small, however, the gig economy is only developing nowadays, coming into its full forth. The positive relationship between the two variables highlights that the market expansion effect dominates the substitution effect in the United Kingdom labor market during the years 2014-2017.

Secondly, the variable median age have a significant correlation with the dependent variable. The 1-year increase in the median age in the region decreases the number of people in employment by 4,14 percent. The paper supposes this is due to the fact that the older the population is, the fewer people would enter the labor force. That could be explained by the fact that the higher is the median age, the greater is the possibility of the decline of the working-age population. For example, that problem exists in South Korea, where the demographic change has triggered the working age-population to lower in absolute terms. (OECD, 2018).

The variable job density also has a significant correlation with the number of people in the dependent employment, the model implies that the increase in the job density in the region by 1 job would increase the dependent employment by 12,8 percent. That figure could seem to be very high, however, it is easily explainable, because an increase in the job density by 1 job would mean that there are more jobs nowadays per one resident of the working-age and there is more possibility for people to find and enter the employment and the labor force.

Moreover, it could be observed that the variable, such as the median annual pay in the region has a positive, but insignificant effect on the dependent employment. Also, the amount of the weekly job-seeker allowance has a negative, but insignificant relationship with the dependent variable. Constant in this regression does not have an important meaning and a direct interpretation.

Overall, the hypothesis stated at the beginning of the paper about the correlation between the gig economy and the dependent employment is not rejected. There is small, but the positive correlation between the gig economy and the dependent employment.

> Heteroscedasticity check

The problem of heteroscedasticity exists when the spread of the residuals is systematically different over the diapason of measured values. Heteroskedasticity might be a problem as its existence makes the invalid the statistical tests of significance. To account for heteroscedasticity, the paper used the function vce (robust), to receive the regression with heteroscedasticity-adjusted standard errors and see, how it differs from the main regression. The results of the regression could be seen in Table 5 in the (6) specification. It could be observed that the results of the regression remain stable, and the influence of the main variable under the investigation remains the same. Consequently, the model does not suffer from the heteroscedasticity problem.

> Variance inflation factor test (VIF)

While regressing the model, it is important to include independent variables that are not correlated with each other to avoid biases. The condition, when several independent variables are highly linearly related to each other is called multicollinearity. Multicollinearity has various adverse influences on the regression model. Firstly, the influence of the dependent variable on the independent variable becomes less accurate. Secondly, standard errors in the presence of multicollinearity become biased - larger than they should be - the possibility of the type II error is increased. Thirdly, in the presence of multicollinearity in the model, small modifications in the analyzed data would lead to large changes in the obtained model.

To check, if there is a collinearity problem in the model or not, one should undertake the variance inflation factor (VIF) test. VIF test results in the index that represent the degree of the increase of the variance of the regression coefficient because of the multicollinearity problem in the model. The results of the VIF test for the model could be seen in Table 7.

***Insert Table here 7***

6. To conclude that there is no multicollinearity in the model, every variance inflation index should be lower than 5. From the results of the table above could be seen that every index of the variable in the regression is lower than 5. It could be concluded that there is no multicollinearity present in the model.

7. Robustness Check

In the further analysis, the number of modifications are undertaken to conduct a robustness check to prove the stability of the achieved results: firstly, the number of different specifications is checked for the first and the second stage of the two-stage least squares model; secondly, the Arellano-Bond model is introduced to prove that the results remain stable after the change in the model.

1. Change in the model specification

> Different specifications of the 1st stage

To consider the robustness checks in the model, the research considers different specifications of the 1st stage of the model: the first specification is the main one with the variables education, internet use, lnGHDI as instruments, the second specification considers adding dummy of the year 2014 into the model, as in this year some changes in the employment law were introduced that might have influenced the introduction of the gig economy.

***Insert Table 4 here***

In the table above in the regression (1) the results of the modified specification are observed. It is seen that the results of the two regression specifications are mainly the same with a slight difference in the degree of the significance under which the null hypothesis of the zero coefficient is accepted.

Under this specification, the 1 percent increase in the annual median annual pay in the region would decrease the number of participants in the gig economy by 0,083 percent, implying the negative correlation similar to the main model without the dummy.

Moreover, quite intuitively, the variable responsible for the percentage of users of the internet has a positive effect in the regression - a 1 percent increase in the number of recent internet users increases the number of gig economy participants by 0,888 percent.

The variable job density also has a negative correlation with the number of participants in the gig economy. The increase in the job density in the region by one point would decrease the number of gig economy participants by 5,3 percent. The effect is similar to the effect in the main regression, with a higher degree of significance.

Another variable that has a significant correlation with the number of participants in the gig economy is the amount of the job allowance paid weekly. The relationship here is significant 33 and implies that when the job-seekers allowance per week is increased by 1 percent, the number of participants in the gig economy is increased by 0,018 percent, implying the positive influence of the variable.

The dummy variable of the year 2014 introduces in this model has a negative influence on the participants in the gig economy, which goes in line with employment rules introduced in 2014.

Overall, it could be observed that the change in the regression specification by adding the dummy variable has not changed the result of the first stage by much.

> Different specifications of the 2nd stage

To provide the robustness check, the paper has also considered different specifications of the 2nd stage. It has taken into consideration adding into the instruments the dummy variable year 2014, using the fixed-effect model to check the results, eliminating the gig economy from the model, and checking the model with fixed and random effects. That was done to prove the stability of results, obtained in the paper.

***Insert Table 5 here***

The regression (2) describes the results of the second stage with adding dummy of the year 2014 to the instruments, the regression (3) checks the results under the random effect model, the regression (4) is the result of eliminating the lngig from the regression under the random effect model, the regression (5) is similar to the regression (4), however, it considers the fixed effect model.

The main moment at which the attention should be paid after the modification is the correlation between the gig economy and the dependent labour.

It might be observed that the main result that was under the consideration in the current paper has not changed by changing the specification of the model: there is a positive correlation between the number of people in the gig economy and the number of employees in the economy. When there is a dummy variable in the model, the 1 percent increase in the number of firms in the gig economy, increases the number of employees by 0,04795 percent. After changing the model to the fixed-effect model, the correlation remains positive, and the 1 percent increase in the number of participants in the gig economy increases the number of employees by 0,07055 percent. The effect is similar to the effect in the main regression. That means, that the main result under consideration is stable.

Considering the effect of other variables after changing the specification, it also seems to be quite stable. The negative correlation between the median age and the number of employees in the economy persists in the model with the dummy variable, the 1-year increase in the median age in the county decreases the number of employees by 4,14 percent, the same effect as in the main regression. In the model (3) with a fixed effect, the 1 percent increase in the median age in the county, decreases the dependent employment by 2,76 percent. Considering the models without the variable responsible for the gig economy the negative correlation of the median age persists. In a random effect model, the negative effect is 10,12 percent, and in the fixed-effect model, the negative effect is 7,82 percent.

The similar significant effect as in the main model is also seen for the variable depicting the job density in the region.

One main difference appears in the regression (4) and (5), it is related to the variable median annual pay in the region that turns out to be significant after eliminating the effect of the gig economy in the regression.

Overall, the regression results are the same, regardless of the specification of the regression that proves the stability of the results of the two-stage least squares model.

2. Change in the model: Arellano-Bond model

Another way to consider the robustness check for the obtained results is to use another model and check the regression results' stability. The paper considers the Arellano-Bond model, which is a dynamic panel data model that allows accounting for the lags of the dependent variable in the regression, estimating the correlations.

***Insert Table 8 here***

The results of the research are presented in Table 8. It is essential to interpret them and compare with the results obtained from the two-stage least squares model.

First of all, it is important to consider the effect of the variable of the number of gig firms in the economy. It could be seen from the model that there is a significant relationship between the number of firms in the gig economy and the dependent labor. One percent increase in the number of gig firms in the economy increases the dependent employment by 0,0429 percent, implying the positive correlation between two variables.

Consequently, from the Arellano-Bond model, it could be concluded that the market expansion effect of the rise of the gig firms in the economy dominates the substitution effect.

As for the lag of the dependent variable, dependent labor, it has a positive and significant effect on the dependent labor. The higher the dependent employment in the previous periods, the higher it is in the current. 1 percent increase in the dependent employment in the previous period, increases the dependent employment by 0,0992 percent in the current. The main assumption of the Arellano-Bond model about the significance of the lag of the dependent variable is satisfied.

Moreover, there is a significant effect of the median age on dependent employment, as in the previous regression models and specifications. The one-year increase in the median age in the region would decrease the dependent employment by 7,4 percent, implying the negative relationship between two variables.

Another variable that in this model turned out to be significant is a value of the job density in the county: the increase in the job density parameter by 1-point increases the dependent employment by 11,25 percent. The amount of the increase in the dependent labor is high as in the labor market there is a significant increase in the number of jobs.

Paying attention to the other variables, it could be noted that the effect of the week-allowance payments to jobseekers and the annual median pay in the region is insignificant due to the model results.

Concluding from the description above, it could be seen that the number of firms in gig employment has a positive effect on dependent employment, and the results remain stable after changing the model. The paper suggests checking the Arellano-Bond model to be important, as it allows us to check the effect of lagged variables on the main dependent variable under consideration.

> Sargan test

Before concluding the validity of the Arellano Bond model, it is important to check the validity of the specification of the model that has been chosen. It could be checked by the Sargan test. Sargan test checks the validity of the overidentifying restriction with the null hypothesis that overidentifying restrictions are valid.

This study identifies the relationship between the gig economy and the dependent labor in the United Kingdom. Digital technologies nowadays are developing at an enormous pace, so, it is very important to count their effect. The gig economy is represented by selling the labor via platforms that use digital technologies for functioning.

That paper contributes to the academic and empirical investigations of the effect of the gig economy sphere, because of the novelty of this topic, there are few investigations connected to the effect of the whole gig industry on the dependent employment. Studies were looking at the effect of the particular gig platform on the dependent employees in the particular industry, such as in the investigation of Farell and Greig (2012). Moreover, the paper contributes to the analysis of the labor market of the country, in which the gig economy nowadays is developing at one of the highest paces.

Gig economy could have two reverse effects on the dependent employment: substitution effects, which affects negatively the dependent employment, and market expansion effect, which affects positively the dependent employment. The question is what effect would be dominating.

The paper has utilized the instrumental variable model of the two-stage least-squares model to solve the problem of endogeneity in the research that could have appeared in the research. The results show that the hypothesis of the correlation between the gig economy and the dependent employment is not rejected and that there is a significant positive correlation between the gig economy and the dependent employment, meaning that the market expansion effect of the gig economy is dominating. The model was checked for the potential problems, such as heteroscedasticity or multicollinearity, and both problems were not confirmed.

Furthermore, the number of robustness check was performed, such as the changes in the specification of the model, the utilization of the Arellano-Bond model, all the robustness checks have shown that the model results remain stable and that there is a positive correlation between the gig economy on the dependent employment, confirming the prevalence of the market expansion effect.

Limitations and directions for further research

As for the limitations of the study in the current paper, the limitations of the datasets could be noted. Considering the pace with which digital technologies are developing and the pace with which new forms of labor are developed, it would be better to use the data of recent years for the analysis. However, these data sets are not available for all the variables that seem to be attractive to include in the regression specification.

Secondly, another limitation is connected to the imperfectness of measuring the size of the gig economy by the number of non-employee firms in the economy. This measure is applicable, as was described before, however, it is not an ideal measure as it could include some misstatements.

The direction for the further research could be to estimate the effect of the gig economy not only on the dependent employment, but also on the wages in the dependent employment, and, after all, to consider the labor policies that could be applied based on the results of the study.

References