do not necessarily reflect the views of UKDiss.com.
As the concern on the evolution of earnings distribution followed by the introduction of new technology escalates in recent decades, it prompts the motivation across both advanced and emerging nations in exploring the significant changes in labour income followed by fluctuation of business cycles in the current economy. With globalization, there is a notably inclination towards skill-biased of technical change (SBTC), which also known as job polarization. It is claimed that there is a positive correlation between in technology development and job polarization over time which implicitly cause an adverse distributional impact on labour market (Acemoglu and Autor 2011, Autor and Dorn 2013, Goos et al. 2014, Michaels et al. 2014). In accordance with the routine-biased technological change (RBTC) hypothesis, there is a rapid relative employment growth for high skilled labour force as well as a modestly increase in employment share for low skilled workers, but relatively negative growth in terms of middle skilled workers employability. The declining share of middle wage occupations is offset by growth in high and low wage occupations.
Nevertheless, the dispersion of employment structure varies across different nations and different industries. Therefore, in this paper I will be investigating whether there is a matching polarization effect between occupational distribution and wage distribution, followed by the introduction of new technologies across the decades. The main interpretation of this research indicates the inclination of replacing labour in routine tasks with recent technological instruments such as computers and machinery resulting an adverse negative impact on middling relative to high-skilled and low-skilled occupations in terms of employment share and labour income. The empirical results found with those at the upper part and lower end of education distribution are more severely affected by higher share of employment vis a vis technological improvement as compared to the median of the wage distribution, further supporting the hypotheses that the role of technology is the main driver of polarization in the U.S., occupational change between skill groups and earnings dispersion across occupations in particular.
Notwithstanding the consistent impact of technology-driven polarization of the labour market, the insufficiency in supporting and explaining the patterns of trends in income polarization across occupations intrigued the interest for further analysis of the connection between the share of employment and occupational wages in any decades. Besides, the lack of reliable sources of exogenous variation that permeates much of this literature, leads to further investigation of employment and wage polarization by skill levels and demographic groups. This analysis highlights three main key features, firstly it assesses the polarization process vis a vis evolution in the composition of the workforce significantly. Secondly, it will provide a clearer insight of the underlying distinct challenges faced within different skill groups in an increasingly polarizing labour market. Thirdly, the variation in workforce composition across occupational skill levels across the time frame is also emphasized in this framework, focusing on periods between post 1990s and post 2000s.
Section 3.1 presents a statistical estimation using econometric strategies to interpret the relationship between employment shares of labour force and real average hourly wages across occupations in one particular industry and thus makes a comparison in empirical results in different decades. Focusing in post 1990s and post 2000s, I would like to investigate the change in employment share to relative wage ratio across these two periods, given the proliferation of technologies in post 1990s and severe economic impact of financial crisis from year 2007-2009 in the labour market. The unobservable endogeneity and limitations such as potential reverse causality effects, economic shocks, consumer preference and empirical challenges faced in running the regression models will be further discussed in Section 3.2 and constructive suggestions will then be raised to tackle the underlying issues. A detailed summary with conclusive discussions through pointing out the plausible solutions and possible demand for future research will be included in the final section of this paper.
2.1 The Canonical Model
The application of canonical model on explaining the impact of wage structure variation and skill differential on income distribution in the labour market is based on the assumption whereby college/high school log wage ratio serves as the abbreviated premium version subject to the relative supply and demand for labour skills. Following the emphasis on skilled and unskilled workers in the canonical model, Acemoglu and Autor (2011) adopt education attainment and skills level interchangeably by unwinding the heterogeneity in skills within education category. The critical elements of the model rely on the factor-augmenting property of technologies which implies that technical change serves as an index of productivity of labour force. The production function can be illustrated as an economy where consumers have utility function defined over two goods producing by skilled and unskilled workers respectively:
) YH = AH H, and YL= AL L
Under a perfect competition condition, all agents in the labour market are set to be price takers. The high skilled and low skilled unit wages are the differentiation of the production function which are given as the values of marginal product of high and low skilled labours :
wH=dY/dH= AHσ-1σ [ ALσ-1σ HL-σ-1σ+AHσ-1σ ]1σ-1
wL= dY/dL=ALσ-1σ[ALσ-1σ +AHσ-1σ HLσ-1σ] 1σ-1
Combining equation (2) and (3), this yields the total skill wage premium paid to high and low skilled labour force which simply reminiscent as the price of skills in the labour market. It is then expressed in logarithm form for simplicity:
lnω =ln(wH/wL) σ – 1/σ ln (AH/AL)-1/σ ln(H/L)
The implication of imperfect substitution between two types of skill level shows that there are not directly skill replacing technologies prone to the shortage in providing an inclusive framework in understanding the evolution of wage distribution and labour development without taking allocation of skills-tasks into account. In addition, the model provides an overview on the uneven income distribution both between and within group. The assumption of difference in supply of efficiency across each workforce with distinct skill levels demonstrates a resemblance between overall income inequality and changes in skill premium in the labour market.
A more general task-based model prior to the canonical model is instead suggested in the paper comprising a framework with continuum tasks, which offers a unique final good categorizing three type of skills: low, medium and high with each labour force endowing with one of these types of skill and supplying efficiency units inelastically. Resembling the Ricardian Trade model, workers in the labour market assumed to have different comparative advantages depending on their education levels and learning abilities which serve as a vital contribution to allocation of skills to tasks. Firms choose the optimal allocation of skills to tasks to achieve a competitive equilibrium in the labour market. A set of tasks is allocated into three sets, one carried out by low skilled, middle skilled and high skilled, assuming the ‘law of one price’ holds whereby same amount of wage being distributed workforce with same skill level performing different tasks in the competitive equilibrium. The relative wage ratio representing the effective relative supply and demand of labour force provides information about the wage structure and earnings inequality across the skill levels, for instance the comparison between middle skilled and low skilled wages:
wM/wL = M AM / L AL =( IH-IL/ IL)(αL (IH )/(αM (IH ))
representing the relative effective supply and demand of middle to low skills. The equilibrium partition of tasks between three types of skill level is illustrated as below:
Supply and Demand
Figure a : Equilibrium allocation of skills to tasks
Nevertheless, it is of paramount importance to note that job polarization does not merely imply a composition of skills available in the labour market but also diversion in allocating skill group across occupations. Introduction of new technologies replacing workers in certain occupational tasks have richer but rather intuitive effect on the earnings distribution and employment patterns in the labour market. In this paper, I aim to further examine the matching of skills to tasks in labour market vis a vis technological development by taking the diversity in earnings distribution for each occupational skill levels, specifically in the U.S. population.
2.2 Literature Review
Goos et. al (2014) on the other hand developed a framework in investigating the job polarization impact through routine biased technological changes (RBTC) and task offshoring hypothesis. The apparent differential effect of job polarisation prior to between-industry and within-industry is justified in the discussion. An industry influenced by routine-biased technological change will result to shifts in occupational employment shares despite the constant effect on output shares. The findings whereby the probability of high paying and low paying occupations has elevated as compared to the middling occupations in each country are consistent with the stylised fact that the downturn in high and low skilled workforce employability is offset by the rise in the share of employment in both high and low skilled occupations as new technologies are introduced. Their results also suggested that the impact of routine biased technological changes (RBTC) tends to outweigh the influence of task offshoring on composition of employment. The empirical evidence shows that industries which heavily focus on routine-based tasks have a lower relative cost of production despite the slight higher level of initial cost of capital for instance, machinery. suggesting that routine based technological change prior to skill-shift plays a vital role between-industry. However, this accuracy and accountability of their findings may be debatable. The decline in employment share of medium levels of skill since 1980s is argued to be driven by skill-biased technological change but not evolution of labour market institution such as contraction in unionization which implicitly related to task offshoring, (Cheremukhin, 2014).
On the contrary, Cheremukhin (2014) highlighted the similarity in job polarization patterns found in the U.S. as well as 16 developed European countries whose labour market institutions structured differently which claimed to play a relative less significant role in polarization expansion. The prominent decrease in routine-based employment rate which represented by middle-skill labour force since 1970 and was fairly uneven during recessions. The huge overall loss in routine-based jobs during the downturn indicates the phenomena so-called jobless recoveries. This scenario seems to be further exacerbated with widespread of new technologies thus contribute to a large shift from non-routine to routine tasks. It is found that the impact of job polarization was not obvious in the 1980s but began in the early of 1990s and intensified in the last decade. The pace of job polarization is unlikely to slow down anytime in the future within the labour market.
The impact on redistribution of income prior to skill-biased technological change (SBTC) accompanied results in an increasingly abundance at the two ends of occupational skill distribution, particularly in both high skilled and low skilled labour force. Employment share within the high skilled occupations was relatively high throughout the past three decades. Despite the economy being in destructive recession, high skilled labour force such as managerial, professional and technical occupations experienced nearly no decline in employment. On the other hand, middle-paid occupations such as sales, office and administrative, production, craft and repair and operators, fabricators and labourers were hit hardest during the Great Recession, with dramatic decline in employment from 7 percent to 17 percent, (Autor, 2010). Therefore, the rapid growth in labour participation in both high and low educational jobs have significantly reduced the employment share of middle skilled workers which hold mostly some college education levels. It is undoubtedly true to say that, the accountability aspect of the classification of skill levels based on labour force education attainment appears to play an indispensable role in assessing the uneven income redistribution in the labour market across decades. Interestingly, The Polarization of Job Opportunities in the U.S. Labour Market paper also highlighted the disparity in polarization process among the sexes in the population. With the rise in feminism awareness, the number of female workers entering managerial, professional and technical field increased significantly, which proportionately decrease the share of employment of female labour force in traditional female occupational jobs such as teaching and nursing. (Autor, 2010) Amid the disparity between the role of international trade and the role of information technology improvement on earnings inequality in labour market remains as a remarkably debatable discussion, economists in general have agreed that the impact of technological shift on change in employability and wage distribution does outweigh the impact of trade in explaining the labour market development. However, the contemporaneous debate between trade and wages interestingly claimed that any labour market impacts due to international trade would be distributed evenly for both low-skilled workers as well as importing-exposed workers. The main attribution to labour market expansion is due to new technological development which complementing high skilled workers and undermining the labour demand, typically in manufacturing sector. Also, in quantifying the effect of international trade on employment share across occupations and the accompanied trade adjustment friction, it is found that the short and medium-run adjustment cost induced by large trade shocks are sizeable entries vis a vis gains from trade. China’s unforeseen bloom with increasing in transformative economic power in emerging market clearly emphasizes the rise of distributional cost. The realization of trade balance for any one country correlating systematically to that for any other country came into a bigger picture, taking US-China relationship as an example whereby industries across US vary widely in their exposure to import competition from China As it is claimed that with the presence of trade openness, it is highly possible that international trade will be closely susceptible to the technical change. Having reviewed all the papers above, it has encouraged the motivation for additional research and findings of the impact of job polarization on the income inequality in the labour market and economic cycle at a wider range of perspectives amid the speedy technological improvement.
Replicating the data achieved in “The Polarization of Job Opportunities in the U.S. Labour Market: Implications for Employments and Earnings.” Center for American Progress and The Hamilton Project, April 2010, it consiststhe information from May/Outgoing Rotation Groups and March U.S. Current Population Survey (CPS), wages are divided and weighted by CPS small weights. Data collected is separated into hourly wages based on the of the logarithm of reported wages for labour forces paid by hour and the logarithm of usual weekly wages divided by hours worked last week for non-hourly workers. The average hourly earnings are calculated using average share of total hours for each group of labour forces over 1980s to 2009s so that investigation will not be affected by the shift in experience, gender composition, regional and race factors of the labour forces. All labour wage paid are deflated by the chain-weighted price deflator for personal consumption expenditures. Whilst, Employment share across different occupations are collected which further categorized using education attainment and median average hourly wage of each labour force.In general, we can classify managerial, professional and technical occupations as high skilled; sales, office and administrative, production, operation occupations as middle skilled and protective services, food preparation and personal care occupations representing low skilled. This orthodox classification and exposition can be explained in the employment classification grid table extracted from Autor’s Hamilton project paper as below.
Figure b : Employment Classification Grid
However, in this analysis I will be extending my finding into two major components, for educational categorization: no high school graduate, high school diploma, some college with no degree, associate degree and bachelor degree are used as the baselines. These occupational-skills classification are divided into three or four major components for in depth and detailed investigation, with education level between no high school and high school diploma are labelled as low skilled; some college with no degree and associate degree known as middle skilled; whereas bachelor degree as high skilled. On the other hand, in terms of median average hourly wage categorization: wage thresholds below 2.577808 are considered as low skilled occupations and those above 2.887567 are considered as high skilled occupations; whereas the wage levels between the two values will implicitly represent as the middling occupations. This classification is based on the information given from the detailed summary statistics of the model.
The analysis examines the variation of employment shares vis a vis technological changes on average real wage of labour force from different occupational skill levels and on the recent growth of U.S. wage differentials. A classic panel data model is used to investigate changes in relative employment shares and wage premium paid for workers by skill levels in the aggregate U.S. labour market, It allows the exploitation of the varying impact that economic cycles have across different occupations and years to estimate the exogenous effect of these cycles on the occupational skill distribution. This model comprises labour force with different skill levels substituting each other, providing a clear explanation and interpretation of the change in income distribution through interaction among occupational skills, job polarization and advancement of technologies.
Despite the accountability aspect and substantial empirical applicability, the model is restricted in providing a clear inclusion of one to one mapping between skills and tasks. The allocation of skills to tasks plays a vital role in relating the impact of technological development on employment and earnings. Also, this model is purely based on the strong exogeneity assumption on technology which may possibly cause empirical challenges in estimation. The first issue is the underlying omitted variables bias prior to unobserved variables, neglected in the analysis which may be correlated with occupational employment shares and act as an influential determinant for average earnings. Furthermore, there is a tendency of omitted bias and reverse causality issue may arise and lead to inaccuracy in the final outcome. The first issue is possible omitted variable bias resulting from unobserved variables which not included in the analysis, such as minimum wage rate in different occupations may be correlated with the regressor, the employment share and act as a determinant of average hourly wage. The second issue is serial correlation between employment share of different skill levels implies that the share of employment of middle skilled workers is likely depend on employment share of both high and low skilled workers. Also, these endogenous effects prone to simultaneity and reverse causality raise the questions, i.e. of how do we ensure that causality is not running both ways and biasing our results?
In this paper, through OLS and Fixed Effect regression models, cluster-robust standard errors are adopted clustered at the occupation level. This allows for general forms of heteroskedasticity and serial correlation within occupations, but still assumes homoskedasticity across different occupations. Allowing for this unobserved cluster effect is necessary for valid inference. Non-robust OLS standard errors would be too small and could lead to incorrect inference. Three regression models are used as below:
- lnwgocct = βlnempocct + αocc+ δαt+εocct
- lnwgocct= βhsDhslnsempocct + βmsDmslnempocct+ βlsDlslnempocct+ αocc + δt + εocct
- lnwgocct= βhswgDhlnsempocct + βmswgDmslnempocct+ βlswgDlslnempocct+ αocc+ δt+εocct
estimating the relationship between the employment shares and average real hourly wages (expressed in logarithm) at different occupational skill levels, whereby
represents the average real wage on hourly basis in occupation
is the log of employment shares for occupation
occupation fixed effects ;
are the time fixed effects and
is the error term. The occupation fixed effects are dummies for each skill levels which control for unobserved occupational characteristics, constant over time but differing between occupations. Time fixed effects are dummies for each year which control for common labour market trends, equal across all occupations but varying from year to year. The inclusion of fixed effect attempts to address endogeneity problems by accounting for unobserved characteristics and trends.
We ran regressions for each model, one regression for each category of the occupational real labour force earnings. Model (1) provides an insight of the marginal effect β across all panel years 1973 to 2009. Model (2) & Model (3) with skill levels dummy variables, gives us the explicit effect of one unit increase in the occupational employment share of each of the three skill levels workers: the high skilled (
), the middle skilled (
) and the low skilled (
) on their average hourly earnings. For each type of skill level, we compute average of the employment shares based on education attainment and average hourly wage level against real labour income over the time period prior to technology development. Taking the credibility of the information collected into consideration, I have included a total of 10 distinct occupations and a time frame from year 1979 to 2009 in the data. The framework comprises of continuum tasks, which in overall contribute to the production of a unique final good with each labour force is endowed with one of three skill levels. Given the variation in prices of different tasks and wage levels for different skill levels in the labour market, firms choose to optimize the allocation of skills to tasks. It is assumed that the productivity of workforces in all tasks can be adjusted in accordance with the technical change.
Figure 1 presents our results from the first regression model, the estimation of the effect of employment share on average hourly wage of labour force across different occupations. The employment share over the change in average hourly wage ratio for each occupation is calculated by dividing employment share of each individual over the total employment shares by average hourly wage over the total of average hourly wages. Since both of the dependent and explanatory variables are represented in log terms, it can be summarized as 1 percentage point increase in the employment share will result in a proportionate 1% increase in average hourly earnings of the labour force. The results notably show the relative greater increase in both high skilled and low skilled labour employment share and earnings as compared to the middling occupations between 1989 to 1999. With the rapid improvement of technologies in the modern world during post 1980s, the replacement of routine-based tasks with computerized equipment reduce the demand for middle skilled labour force by a significant amount. Nevertheless, our second main finding indicates that followed by the massive recession and financial crises from year 2007 to 2009, all occupational employment shares and average hourly wages slowdown substantially, even for high skilled and low skilled occupations, specifically managers and protective services. This is mainly due to the lack of demand in labour force across different industries which prone to higher unemployment rate which grow in accordance with the employment share and labour income. Thirdly, most of the middle skilled cognitive and production activities, such as bookkeeping, clerical work, and repetitive production work are represented as routine-based tasks and experienced negative change in employment share, excluding sales sector. On the contrary, with the introduction of new technologies these routine-based tasks are increasingly substituted with computer software and carried out by machinery, thus leading to a greater task-shift towards either abstract or manual tasks. which lie at the two opposite ends occupation-skill distribution. This dramatic task-shift process causes a severe change in employment share and average hourly wage paid across occupations, benefiting the labour forces which hold a comparative advantage in non-routine tasks and worsening the employability of labour force which hold a comparative advantage in routine-based tasks. This uneven employment share-wage paid distribution will eventually result in an adverse effect on income inequality in the US population. The intensive discussion of the income inequality prior to skill-biased technical change (SBTC) within the labour market prone to a deeper insight and understanding of the employability as well as polarization within the education distribution and wage distribution which then demonstrated in regression model (2) and (3).
Table 1 shows the results of regression models, the positive estimated coefficient of 0.0690 illustrates the direct correlation between employment share and real labour income of an occupation. It shows the coefficients on share of occupational employment are statistically significantly different from all categories. With a rise in labour participation in a particular occupation, the real wage of the labour force will increase by 6.9%. Also, it demonstrates the credibility and consistency of employment share estimation based on education attainment of labour forces as compared to the median average hourly wage which claimed to be rather too comprehensive and lead to inaccuracy in data results and will be discuss more in detail in the rest of the paper. Further exploring the relationship between the share of employment and real labour income with the introduction of new technologies over time, I have discovered that there is a relatively expansion of employment share in high skilled occupations vis a vis low skilled and middle skilled occupations. The -0.0092 negative correlation within middle skilled labour force reiterated the fact that middling occupations are most affected by the change in composition of labour force. In other words, the demand elasticity of middle skilled workers is fairly elastic as it highly depends on the introduction of new technologies which will prone to excess supply of this range of labour force and reduce the real wage level to a great extent. Implicitly, this forces workers with ‘some college’ education level to either further enhancing their skill levels through pursuing higher education attainment or shifting to lower wage paid job tasks. Favouring towards high skilled and low skilled occupations with substantial contraction in job opportunities for middling occupations. The wage paid to high skilled professional, technical and managerial occupations and low-education food services, personal care and protective service occupations rises concretely. On the contrary, the average earnings of white collar clerical, administrative and sales occupations and blue-collar production, craft and operative occupations decline proportionately with an excess supply of labour forces in this sector. Besides, in order to minimise the underlying heteroskedasticity and serial correlation issues in the regression models, cluster-robust standard errors will be used, by allowing the general forms of heteroskedasticity and non-zero serial correlation within occupations, at the same time homoskedasticity between occupations still holds. The inaccurate inference due to non-robust OLS standard errors is therefore tackled and solved. By allowing the serial correlation between the variables, employment share across three different occupational skill levels in the fixed effect analysis, the higher accuracy and consistency findings are obtained.
Table 1: Data Analysis
Figure 2 illustrates the employment share to relative wage ratio of different occupational skill levels across the period from year 1979 to 2009. I have calculated the ratio by dividing the employment share of each workers with three major types of occupational skill levels over their relative average hourly wage to demonstrate a more concise and accurate estimation. On the other hand, Figure 3 based on the median wage classification explains the correlation between change in employment share and relative average hourly wage of each occupational-skilled workers through presenting the wage level distribution. Also, the graphs are represented using a two-way scatter methodology so that the distribution based on education attainment and median wage can be shown in a clearer manner. Comparing the results both of the figures, I have concluded that the estimation based on education attainment seems to be a better and logical method in classifying the occupational skill levels. This is due to a relatively lower volatility in employment share to relative wage distribution and more stable fitted value. In general, representation of an upward sloping fitted line shows that the positive correlation of employment share and average hourly wage within labour market across the time frame.
Importantly, this model allows for new technologies that may explicitly cause a significant skill-biased technical change (SBTC) in the labour market, notably a more severe impact on middling occupations. Figure 4 provides an overview of trends in occupational composition with respect to the change in employment share and average hourly earnings at middle-skilled level in the U.S. labour market development over the decades.
Based on the panel data distribution graph above, even though the advancement of technology does not constitute any substantial change in average hourly earnings of the middle-skilled workers, but it is important to note that the skill-biased technical change which implicitly leads to job polarization cause a dramatic fluctuations in the share of employment within this skill level, as it erodes the comparative advantage and reposition them from some of the tasks they were previously performing. It is predicted that this will cause a detrimental effect on the wage paid on the middling occupations in the long run and thus exacerbate the income inequality in the U.S. labour market development.
Extending the specification into investigating the initial regression model with interaction with two distinct major time period which are post 1990s and post 2000s, it is clear to note that the effect job polarization on employment share and average hourly earnings of skill-based occupations is more severe during post-recession and revolution of technology, middle skilled workers in particular. A theoretical discussion is then provided for these stylized facts. The speculation presumes in the post 2000s, followed by the occurrence of Great Recession and Financial Crisis from 2007 to 2009 has declined the employability in the U.S labour market by a great amount. As the unemployment rate across different occupations and industries increase dramatically, labour forces with lower qualifications also known as having lower education levels tend to lose their jobs substantially. This enforced the statement whereby job polarization is further intensified across the decades. The advancement of information technology (IT) does not cause a negative impact on abstract, analytical tasks specialized by high skilled workers but rather promote a greater productivity. Nevertheless, this has led an adverse effect on routine -based tasks which are normally conducted by the middle-skilled labour force particularly assembly-line employability whereby a set of instructions which a machine can easily follow. In a nutshell, the fastest growth in demand for the most educated workers and the sharpest decline in demands for people with intermediate levels of education was particularly obvious for industry with higher adoption of new technologies. However, the odd findings based on the fixed effect results in post 2000s show that despite the relative smaller change in the composition in the labour market during this period, the low skilled workers are affected greatly as compared to the other two occupational skill levels, with a -0.00686 estimated coefficient. Further investigation is hence conducted to target this issue.
Hypothesis testing on whether the post 1990s coefficients were statistically different from those for the post 2000s period is carried out through regressing the average hourly wage on employment share of middling occupations specifically, including the interactions with year post 1990s and year post 2000s. Controlling the year effect across this panel data model and adopting the standard robust errors for higher accuracy, the results indicate that the p value differs from zero to a great extent, 0.6686 which implies that there a corresponding strong evidence to reject the null hypothesis. It shows that the marginal effect of middle skilled workers’ employment share on their average hourly earnings in year post 1990s differs from the marginal effect in year 2000s. A plausible evidence supporting the main argument of this paper is hence provided, illustrating the distinct impact of change in workforce composition between year post 1990s and year post 2000s. Another notable finding from this empirical result is the considerably lower value in the R-squared interpretation of the linear regression model. There might be a possibly whereby such small value of R-squared estimation, 0.0696 may invoke a misleading investigation. However, I would contend that valuing the credibility of the results based on R-squared value seems to be rather too subjective since it is claimed to be a handy and seemingly intuitive measurement on how effective the regression model fits the observable data sets. Despite the fact that R-squared gives an estimate of how closely related the relationship between the regression model and the response variables, it does not provide a formal hypothesis test for this correlation. Adjusted R-squared should therefore be calculated, also to mitigate the concern of invalid exogeneity assumption and address the attenuation bias from measurement error which may possibly arise with fixed effects panel model, the difference-in-difference estimation in the context of the linear regression model will be conducted as an effective method to counter this issue.
3.2.1 Empirical Analysis
The job polarization may not occur evenly across decades, the “hollowing out” effect of the occupational skill distribution has distinct consequences in different time periods. In this section I document how sensitive the main findings are subject to the changes in two major periods which are year post 1990s and year post 2000s. Given the arise of endogeneity issue, I further extend the empirical analysis using Difference-in-difference approach to generate some exogenous variation in explanatory variable, employment share so that the responses of dependent variable, average hourly wage can be interpreted as causal. Through exploiting the difference in trends between treated (post 2000s) period and control (post 1990s) period over time focusing on middling occupations. This is mainly because with the improvement in technologies which prone to a large movement of quotidian tasks being replaced by computerized devices, the rise in difficulty of programming the complex technological development clearly explains why there is relatively less comparative advantage of middle class workers for the past couple of decades. The DID estimation is based on conditions on group level instead of individual level effect. This sensitivity analysis is selected as a measurement to resolve the endogenous problem and identify the effect of interest, fundamentally based on the potential outcome model through specifying the treatment, outcome and treatment effect recast in context of a linear regression model. In order to simplify the presentation, we will consider a binary treatment whereby post 2000s =1 and non-post 2000s =0, by restricting the time frame from 1990 to 2009 when running the regression model. Here, the comparison in outcomes of average hourly wage between treated and control periods is observed by assuming the time trends in outcomes is the same for the treated and control groups, whereby employment time trends are the same in two main periods,
- lnwgooct = = β1Dmslnemp*Dpost2000socct +uocct
represents the real average hourly wages in a given category of the skill distribution, in occupation
is the share of employment of labour force for occupation
represents the dummy variable of skill levels with
indicating middling occupations whereas,
indicating non-middling occupations.
represents the dummy variable with
as treatment period and
as the control period.
Nevertheless, the final results show that the change in average hourly wage in year post 1990s and post 2000s documented above finds a clear irreconcilable in the marked divergence in educational earnings among middle skilled labour force in different duration. In the early 1990s, labour market has been moving towards a world where employment and job prospects are directly related to the technological based tasks. The share of occupational employment is mostly likely to increase proportionately with the average hourly wage across different skill levels in a stable motion. Whereas, the eruption of Great Recession and occurrence of financial turmoil starting in year 2007 induced the substantial changes in the relationship between employment share and average hourly earnings across occupational skills due to the fluctuations in business cycle. Figure 5 provides a concise illustration of the interpretation by plotting a two-way graph of Difference in Difference (DID) estimation between the control period post 1990s and treatment period year post 2000s. The gap between the red scatter points from year 2000 to year 2009 indicates the variation in the direction of growth with respect to employment share and average hourly earnings of the labour forces with distinct skill levels. In summary, the disparity in the movement in labour participation across occupation with different skill levels intensify the change in workforce composition from year 2000 onwards and thus create an uneven distribution in the individual income in the U.S. labour market.
However, these contributions should not be taken as the only causation of income inequality in the labour market, followed by the recent slowdown in educational attainment which has been particularly severe within male as compared to female young adults, prone to an extended sharp rise in the uneven distribution of earnings. The respective decline in the male education attainment leading to a great slump in the male labour force participation across different industries cause a stagnation in labour earnings. This interesting finding can be possibly explained as a significant impact due to the rise in feminism awareness accompany with a greater emphasis and encouragement in woman leadership and involvement at higher education level in the 21st century and thus influence labour participation in different skill levels occupations. With the rise in number of female workforce taking up professional courses which are mathematical, technical and engineering related will certainly cause a major disruption on data computation in this framework. The ignorance of female labour force participation when collecting data is likely to be counted as one of the limitations of this model. In addition, one can also argue that the degree of technology adoption across different industries and different nations may also cause distinctive results and interpretations in this framework. It is argued that developed countries such as United States, United Kingdom, Japan have relatively a greater adoption and higher diffusion of advancement of technology as compared to developing countries such as Indonesia, South Korea, Turkey. The skill-biased technical change(SBTC) effect on income inequality followed by a better grip in the new technologies in developed countries, United States in this case is particularly more severe and dramatic. Also, there is an emerging concern on the close relationship between international trade and technological change and its significant impact on wage component across occupations and industries in recent decades. Grossman and Helpman (1991) assert the fact that the long run growth of an economy relies on the availability of the resources which comprise of the human and capital abundance subject to the index of technologies which play an indispensable role in determining productivity, wage structure and economy development. A higher level of trade openness is often accompanied with a greater amount of technology adoption in order to increase its productivity to reach the targeted output level as it implicitly encourage firms to develop and adopt labour-saving technologies. The intuition is rather simple implying trade skill-biased technical change will be induced with the attempt to accelerate the relative productivity of skilled workers and thus indirectly worsen the employability of unskilled workers. Trade liberalization could promote skill-biased technical change and generate wage inequality more than predicted by standard trade theory, provided skilled and unskilled workers substitute each other. Nevertheless, these statements tend to questionable as the effect of trade on technology may not be a plausible explanation on the uneven wage distribution in the labour market as it depends heavily on the operational price effect and the relative price changes in the market is likely to be too small.
3.2.2 Robustness Check
As numerous of underlying endogenous effects exist, I have adopted a few functional measurements to improve the validity of this analysis. First of all, a panel data is used to investigate the relationship between average hourly earnings and employment share of labour force at different skill levels, controlling the factors that differs across occupations but do not vary over time, unobservable variables which cannot be included in the regression and excluding the omitted variable bias when it is constant over time across a given occupation. As mentioned above, the data collected on real wages and employment share of the labour force are expressed in logarithm form to narrow down the heterogeneous shift in the labour market. Instead of using the standard error, I have adopted the robust standard error when running the regression models, so that any overestimation and underestimation of the variability in estimation can be avoided through further strengthen the asymptotical inference and homoskedasticity assumption while allowing possibility of heteroskedasticity. Since, the plausibility of regression coefficients is rather too inexplicit, I have conducted a Hausman test to observe any apparent disparity among the estimators which may possibly show a potential misspecification so that an attribution of causal effects is permitted.
In this paper, I have discovered the influential impact of skill-biased technical change on employment and wage distribution prior to introduction of new technologies and its consecutive significant changes on income inequality in the labour market. A noteworthy inference of this analysis is that technical change result in favoritism in a particular skill level of workforces will reduce the real wage and employment share of another group. It can be concluded that, following the substitution of routine-based tasks with computerized devices, the real wage and employment share of middling occupations are severely affected, leading to substantial shift to high skilled and low skilled job specialization. The variation of workforce composition due to job polarization differs in two eventful periods which are during the post 1990s with the introduction of new technologies and post 2000s with the occurrence of Great Recession and Financial Maelstrom in Year 2007 to 2009. The significant implication of how to classify the skill levels of labour forces and its distributional impact on occupational wage and employment arrangement is also emphasized. Nevertheless, I have come into a conclusion that in spite of the prominent findings from this research, there are a few legitimate findings which would be beneficial for further investigation in the uneven income distribution in the labour market across different nations globally. Firstly, to which extent the technical change has evolved and its degree of adoption in various countries have a vital role determining the level of income inequality. Whether the technological variation exhibits a state dependence effect and whether this effect affects various factors such as allocation of tasks, classification of labour skill levels, task routineness intensity differentially is ultimately an empirical question. In addition, the elasticity of substitution between skill levels modulates how influential these effects are, for instance its indication for how technical change and changes in wage levels respond to the change in relative demand and supply of goods and labour force participation. In summary, these potential factors can be viewed as promising topics for future research to understand the consequence of income heterogeneity in a broader and extensive aspects. The policies such as employment quota system and social safety net are suggested to be regulated considerably to protect those most vulnerable to the economic cycles and mitigate the extreme income inequality condition.