DETERMINANTS OF UNEMPLOYMENT AND EARNINGS IN SOUTH AFRICA. Master of Science in Statistics

DETERMINANTS OF UNEMPLOYMENT AND EARNINGS IN SOUTH AFRICA Master of Scence n Statstcs I.N Mathebula 2017

DETERMINANTS OF UNEMPLOYMENT AND EARNINGS IN SOUTH AFRICA by Inocent Nelson Mathebula RESEARCH DISSERTATION Submtted n fulflment of the requrements for the degree of Master of Scence n Statstcs n the FACULTY OF SCIENCE AND AGRICULTURE (School of Mathematcal and Computer Scences) at the UNIVERSITY OF LIMPOPO SUPERVISOR: CO-SUPERVISOR: Prof A Tessera Mr. N Ybas 2017

DEDICATION I would lke to dedcate the success of ths dssertaton to everyone who supported me throughout the duraton of complng ths research work.

DECLARATION I declare that DETERMINANTS OF UNEMPLOYMENT AND EARNINGS IN SOUTH AFRICA s my own work and that all the sources that I have used or quoted have been ndcated and acknowledged by means of complete references and that ths work has not been submtted before for any other degree at any other nsttuton. Inocent Nelson Mathebula 07 July 2017 Full names Date

ACKNOWLEDGEMENTS I want to thank the followng for ther respectve contrbutons to ths dssertaton: God for gvng me the strength, patence, and wsdom to carry ths work My wfe, Tebogo Mathebula, for her uncondtonal love, support and encouragement. My chldren, Accord and Khenswe Mathebula, for ther support and understandng. My parents, Tsakan, Joshephna and Gezan Mathebula, for standng by me throughout ths journey. My brothers, Wlly, Oths, Smon, Ncholas, Lawrence, Hlulan and Matmba, Not forgettng my ssters, Tnyko, Zelda, Mhlot, and Sbongle for beng there for me. A specal thank you to my supervsor, Prof. A. Tessera, for hs gudance, support and encouragement. He was there to keep me on the race even durng the tmes when I wanted to gve up. My jont supervsor, Mr. N. Ybas, for hs support and gudance. Small Gyan Socal Club for ther support and understandng when I could not be wth them due to ths work. Ms. Mmokela Choeu for her support and sacrfces. The Stols famly for ther support and warm welcome n ther home. Mr. Phllemon Dkgale for hs carng and avalablty to assst me. Lastly, Mr. Mokgoropo Makgaba for hs advces, and support durng the way.

ABSTRACT South Afrca s one of the countres wth chronc hgh unemployment rate. The unemployment rate has consstently been above 24% for a consderable perod of tme. It s mportant for polcy and decson makers to know the type of persons who are unemployed, and underemployed n order to come up wth the rght nterventon. The purpose of ths study was to fnd and descrbe the determnants of unemployment, underemployment, and earnngs n South Afrca. In order to realze the objectves of the study, secondary data from 2012 Quarterly Labour Force Survey was used. Statstcs South Afrca collects labour market related nformaton from persons between the age of 15 and 64. The data have nformaton on status of unemployment, underemployment and earnngs and other related to varables. Logstc regresson was appled on the data and t was found that age, gender, populaton group, martal status, level of educaton, and provnce were sgnfcant determnants of unemployment n South Afrca. Gender, populaton group, sector, martal status and contract duraton were found to be sgnfcantly assocated wth tmerelated underemployment. Generalsed lnear model was appled on the data and t was found that gender, populaton group, martal status, level of educaton contract duraton, geographcal locaton, and sector were the determnants of earnngs. KEY CONCEPTS Logstc regresson; Generalsed Lnear Model; Unemployment; Gender; Age; Populaton group; Martal status, Hghest educatonal level completed; contract duraton, sector, unon membershp, Geographcal locaton and Provnce. v

TABLE OF CONTENTS DEDICATION... DECLARATION... ACKNOWLEDGEMENTS... ABSTRACT... v LIST OF TABLES... v LIST OF FIGURES... x CHAPTER ONE: INTRODUCTION AND BACKGROUND... 1 1.1 INTRODUCTION... 1 1.2 RESEARCH PROBLEM... 2 1.3 LITERATURE REVIEW... 2 1.4 PURPOSE OF THE STUDY... 2 1.5 RESEARCH METHODOLOGY... 3 1.6 SIGNIFICANCE OF THE STUDY... 3 CHAPTERT TWO: LITERATURE REVIEW... 4 2.1 INTRODUCTION... 4 2.2 COMPARISON OF DEFINITIONS... 4 2.2.1 Internatonal Perspectve... 4 2.2.2 South Afrcan Perspectve... 5 2.2.3 Smlartes between global and South Afrcan s perspectve... 6 2.3 ESTIMATES OF THE INDICATORS... 6 2.4 STUDIES ON FACTORS THAT INFLUENCE UNEMPLOYMENT... 7 2.4.1 Gender... 7 2.4.2 Age... 8 2.4.3 Populaton group... 8 2.4.4 Martal status... 9 2.4.6 Geographc locaton... 11 2.4.7 Concluson... 11 2.5 STUDIES ON FACTORS THAT INFLUENCE TIME-RELATED UNDEREMPLOYMENT... 12 2.5.1 Gender... 12 2.5.2 Age... 13 2.5.3 Populaton group... 14 2.5.4 Martal Status... 14 2.5.5 Level of educaton... 15 v

2.5.6 Geographc locaton... 16 2.5.7 Sector... 16 2.5.8 Contract duraton... 17 2.5.9 Concluson... 17 2.6 STUDIES ON FACTORS THAT INFLUENCE EARNINGS... 17 2.6.1 Gender... 17 2.6.2 Age... 18 2.6.3 Populaton group... 19 2.6.4 Martal status... 20 2.6.5 Level of educaton... 21 2.6.6 Geographc locaton... 22 2.6.7 Unon membershp... 22 2.6.8 Concluson... 23 CHAPTER THREE: RESEARCH METHODOLOGY... 24 3.1 INTRODUCTION... 24 3.2 THE DATA... 24 3.3 STATISTICAL METHODOLOGIES USED... 25 3.3.1 Logstc regresson... 25 3.3.2 Estmaton of model parameters and standard errors... 25 3.3.3 Evaluaton of the ftted model... 26 3.3.4 Interpretaton of coeffcents... 27 3.3.5 Model buldng or varable selecton for logstc regresson... 28 3.4 Generalsed lnear model... 28 3.4.1 Three components of Generalsed lnear model... 29 3.4.2 Estmaton of the coeffcents... 31 3.4.3 Testng for the sgnfcance of the coeffcents... 33 3.4.4. Evaluaton of the ftted model... 33 CHAPTER FOUR: DATA ANALYSIS AND FINDINGS... 35 4.1 INTRODUCTION... 35 4.2 THE DATA... 35 4.3 UNEMPLOYMENT... 36 4.3.1 Introducton to the secton... 36 4.3.2 Descrpton of the data... 36 4.3.3 Interacton between the explanatory varables and response varable... 39 4.3.4 Applcaton of the logstc regresson analyss... 41 v

4.4 TIME-RELATED UNDEREMPLOYMENT... 45 4.4.1 Introducton... 45 4.4.2 The descrpton of the data... 45 4.4.3: The nteracton between the explanatory varables and response varable... 48 4.4.4 The applcaton of logstc regresson model... 51 4.5 EARNINGS... 53 4.5.1 Introducton... 53 4.5.2 Descrpton of the data... 53 4.5.4 The applcaton of the Generalsed lnear modellng... 59 CHAPTER FIVE: DISCUSSION OF RESULTS AND RECOMMENDATION... 62 5.1 Introducton... 62 5.2 Dscusson of the results and concluson... 62 5.2.1 UNEMPLOYMENT... 62 5.2.2 TIME-RELATED UNDEREMPLOYMENT... 65 5.2.3 EARNINGS... 67 5.3 Recommendatons... 68 5.4 Lmtatons of the study... 69 REFERENCE... 70 APPENDIX... 76 v

LIST OF TABLES Table 4.1: Classfcaton by unemployment status and age group, gender and populaton group... 37 Table 4.2: Classfcaton by unemployment status and martal status and level of educaton... 38 Table 4.3: Classfcaton by unemployment status and geographc locaton and provnce... 39 Table 4.4: Classfcaton of persons by gender, age group and unemployment status... 40 Table 4.5: Classfcaton of persons by gender, populaton group and unemployment status... 40 Table 4.6: Classfcaton of persons by age group, level of educaton and unemployment status... 41 Table 4.7: Classfcaton Table... 41 Table 4.8: Omnbus Tests of Model Coeffcents... 42 Table 4.9: The summary table... 42 Table 4.10: Hosmer and Lemeshow Test... 43 Table 4.11: Logstc regresson analyss for the determnants of unemployment... 44 Table 4.12: Classfcaton by underemployment status and age group, gender and populaton group 46 Table 4.13: Classfcaton by underemployment status and level of educaton and martal status... 47 Table 4.14: Classfcaton by underemployment status and geographc locaton and provnce... 47 Table 4.15: Classfcaton by underemployment status and sector, contract duraton and unon membershp.... 48 Table 4.16: Classfcaton of persons by gender, level of educaton and underemployment status... 49 Table 4.17: Classfcaton of persons by gender, populaton group and underemployment status... 49 Table 4.18: Classfcaton of persons by gender, sector and underemployment status... 50 Table 4.19: Classfcaton of persons by gender, contract duraton and underemployment status... 51 Table 4.20: The classfcaton table of the response varable... 51 Table 4.21: Model summary of the model... 52 Table 4.22: Hosmer and Lemeshow test... 52 Table 4.23: Logstc regresson analyss for the determnants of underemployment... 53 Table 4.24: Mean monthly earnngs by age groups, gender and populaton groups... 54 Table 4.25: Mean monthly earnngs by Martal Status and Educaton... 55 Table 4.26: Mean monthly earnngs by sector, unon membershp, and contract duraton... 56 Table 4.27: Mean monthly earnngs by sector, unon membershp, and contract duraton... 56 Table 4.28: Classfcaton by earnngs and gender, and populaton group... 57 Table 4.29: Classfcaton by earnngs and gender, and level of educaton... 58 Table 4.30: Classfcaton by earnngs and gender, and Martal Status... 58 Table 4.31: Mean, standard devaton and coeffcent of varaton by earnngs group.... 60 Table 4.32: GLM coeffcents and p-values.... 61 Table A1: Tests of condtonal ndependence for gender and the status of unemployment after adjustng for age group... 76 Table A2: Tests of condtonal ndependence for gender and the status of unemployment after adjustng for populaton group... 76 Table A3: Tests of condtonal ndependence for age group and the status of unemployment after adjustng for level of educaton... 76 Table A4: Tests of condtonal ndependence for gender and the status of underemployment after adjustng for level of educaton... 76 Table A5: Tests of condtonal ndependence for gender and the status of underemployment after adjustng for populaton group... 77 Table A6: Tests of condtonal ndependence for gender and the status of underemployment after adjustng for sector... 77 Table A7: Tests of condtonal ndependence for gender and the status of underemployment after adjustng for contract duraton... 77 v

LIST OF FIGURES Fgure 4.1: Spke plot on Earnngs.. 60 Fgure A1: The dstrbuton of monthly earnngs by age groups... 77 Fgure A2:The dstrbuton of monthly earnngs by gender... 78 Fgure A3: The dstrbuton of monthly earnngs by populaton groups... 78 Fgure A4: The dstrbuton of monthly earnngs by age groups... 79 Fgure A5: The dstrbuton of monthly earnngs by educaton... 79 Fgure A6: The dstrbuton of monthly earnngs by sector... 80 Fgure A7: The dstrbuton of monthly earnngs by contract duraton... 80 Fgure A8: The dstrbuton of monthly earnngs by unon membershp... 81 Fgure A9: The dstrbuton of monthly earnngs by geographcal settlement... 81 Fgure A10: The dstrbuton of monthly earnngs by provnce... 82 x

CHAPTER ONE: INTRODUCTION AND BACKGROUND 1.1 INTRODUCTION Most emergng economes, ncludng South Afrca, suffer from persstent and long term hgh levels of unemployment. What makes South Afrca s experence dfferent from the others s that the rate s very hgh and ncreasng. Accordng to a recent report by the Organsaton for Economc Co-operaton and Development (OECD, 2014), the long-term unemployment rates for Brazl, Indonesa and Turkey are just a bt lower than ten percent and that of South Afrca s around 24%. The latest fgures show that the unemployment rates for the frst three countres have decreased whereas the rate for South Afrca has ncreased. The number of unemployed persons n South Afrca ncreased by 824 000 from 4.3 mllon n 2008 to 5.1 mllon persons n 2014 whle employment to populaton rato declned by 3.1 percentage ponts from 45.9% n 2008 to 42.8% n 2014 (Statstcs South Afrca, 2014). The unemployment rate n South Afrca vares by gender, populaton groups, age groups, levels of educaton, and other factors. Accordng to Statstcs South Afrca (2014), the unemployment rate for women s 27.2% and 23.3% for men. The unemployment rate for black Afrcan s four tmes that for whte populaton; and about three tmes that for Indan/Asan populaton. The youth (15 34 years) and those wth secondary educaton have the hghest unemployed persons compared to other age groups and levels of educaton respectvely. In addton to the hgh level of unemployment, the South Afrcan economy s also characterzed by a substantal varaton of ncome dstrbuton across gender, populaton groups, educatonal level and the lke. Accordng to Statstcs South Afrca (2014) the medan monthly ncome for men were R3 500 and for women R2 600. Black Afrcans earned 28% of what the whte populaton earned; 47% of what Indans/Asans earned, and 92% of what the coloured populaton earned. Smlar substantal varatons are observed by age and educatonal level. 1

Two of the challenges facng South Afrca are reducng the unemployment level and creatng favourable condtons for lowerng the sgnfcant and unfar varaton n ncome dstrbuton. In ths regard, a study on determnants of unemployment and earnngs could be useful. 1.2 RESEARCH PROBLEM South Afrca s faced wth hgh levels of unemployment, ncreasng tme-related underemployment, and a severe and unfar ncome dstrbuton. To reduce the unemployment rate and narrow the ncome gap, t s mportant to study the factors related to the varable of nterest. The study analyses Quarterly Labour Force Survey (QLFS) data collected between July and September 2012. Statstcal methodologes such as logstc regresson and generalsed lnear model wll be used to determne and descrbe the determnants of unemployment, tme-related underemployment and earnngs n South Afrca. 1.3 LITERATURE REVIEW Several researchers around the world have conducted smlar studes to determne and descrbe the determnants of unemployment, tme-related underemployment and earnngs. In order to acheve ther objectves, they used dfferent statstcal methodologes such as logstc regresson, survval analyss, cross sectonal and probt regresson analyss. Varables such as gender, age, populaton group, martal status, level of educaton, geographcal locaton, sector, contract type, and unon membershp were found to be strongly related to unemployment, tme-related underemployment and earnngs. 1.4 PURPOSE OF THE STUDY The am of the study s to nvestgate and descrbe the determnants of unemployment, tme-related underemployment and earnngs n South Afrca. The fndngs of the study can be used to better understand and descrbe factors related to hgh unemployment rate and severe ncome gap. 2

The study s objectves are as follows: To determne factors assocated wth unemployment n South Afrca; To dentfy factors assocated wth tme-related underemployment; To dentfy the determnants of earnngs; and To gve recommendatons that could help n reducng unemployment and narrowng the ncome gap. 1.5 RESEARCH METHODOLOGY The Quarterly Labour Force Survey(QLFS) 2012 data wll be used to better understand and descrbe the determnants of unemployment, tme-related underemployment and earnngs. The study wll be restrcted to persons between the age of 15 and 64 years wthn the borders of South Afrca. Cross tabulaton, ch-square tests, odds ratos, logstc regresson and generalsed lnear model wll be appled on the data n order to determne the determnants of unemployment, tme-related underemployment, and earnngs. Statstcal Package for Socal Scence (SPSS) wll be used to carry out data analyss. 1.6 SIGNIFICANCE OF THE STUDY The study seeks to establsh factors that are lnked to unemployment, tme-related underemployment, and earnngs. The result of the study may help polcy makers n formulatng strateges and nterventons that may help n lowerng unemployment rate and ncome gap n dfferent sectors of the socety. 3

CHAPTERT TWO: LITERATURE REVIEW 2.1 INTRODUCTION Ths chapter gves an overvew of studes conducted by researchers on three types of labour ndcators; vz. unemployment, tme-related underemployment, and earnngs. Frst the defntons of the terms are gven, n the subsequent estmates on ndcators, and lastly, sectons studes made on factors that nfluence the ndcators are revewed. 2.2 COMPARISON OF DEFINITIONS 2.2.1 Internatonal Perspectve The Internatonal Labour Organsaton (ILO) defnes the unemployed as all persons n the workng age populaton who durng the reference perod were wthout work, avalable for work and seekng work (ILO, 2003). The excluson of the workng age populaton threshold makes nternatonal comparson dffcult. Tme-related underemployment relates to employed persons who were avalable and wllng to work addtonal hours, and had worked less than a threshold workng tme durng the reference perod (ILO, 2003). Ths defnton provdes countres wth a chance to decde on tme-related underemployment threshold workng tme. Ths allows statstcal organsaton or ndvduals to measure accordng to the country set tme or legslature. However, the threshold workng tme can lead to underreportng or over-reportng of the tme-related underemployment. Earnngs relates to remuneraton n cash and n knd pad to employees, as a rule at regular ntervals, for tme worked or work done together wth remuneraton for tme not worked, such as annual vacaton, other pad leave or holdays (ILO, 1973). The defnton permts organsatons to decde on what can be used to estmate or determne employee s actual ncome. However, t does not nclude the amount or remuneraton pad to employees who spent tme away from ther regular places due 4

to work and also the remuneraton of employees workng n the ndustres where they receve tps from customers. 2.2.2 South Afrcan Perspectve Unemployment refers to those persons, aged 15 to 64 years, who were wthout work n the reference week, actvely looked for work or tred to start a busness n the four weeks precedng the survey ntervew and would have been able to start work or would have started a busness n the reference week (Statstcs South Afrca, 2008). The defnton provdes the workng age populaton or age cut-off of ndvduals who are economcally actve and can be used to estmate unemployment rate. However, there are persons who are aged 65 years and older who can be classfed ether as employed or unemployed. Ths leads to underreportng or over-reportng of the country s unemployment rate. Tme-related underemployment refers to those persons who worked less than 35 hours n the reference week and were avalable to work addtonal hours (Statstcs South Afrca, 2008). It provdes the condtons used to classfy employed persons as underemployed. However, It can only be used for natonal and nternatonal comparson wth countres usng smlar condtons. Earnngs refer to the amount an employer pays an employee for work done. It s a fxed ncome for servces, whch s usually pad on a weekly, b-weekly or monthly bass (Statstcs South Afrca, 2008). It does not only nclude one payment reference perod, but dfferent payments dfferent perods such as weekly, b-weekly or monthly. Ths can be used to compare wth other regons at a dfferent earnng perod. The defnton measure earnngs as an amount earned from an employer only. It does not nclude mones earned by tps and bonuses. 5

2.2.3 Smlartes between global and South Afrcan s perspectve The ILO and Statstcs South Afrca use smlar concepts and defntons. The only dfference that the ILO uses general standards that are sutable for all countres, whle Statstcs South Afrca usng concepts and defntons sutable for South Afrcan stuaton. 2.3 ESTIMATES OF THE INDICATORS 2.3.1 Unemployment Unemployment s experenced by both developed and developng countres and regons n the whole world. ILO (2015) reported that there were 201.3 mllon unemployed persons n 2014 n the whole world, of whch 74 mllon were youth (15 24 years). The unemployment rate for the youth was almost three tmes hgher than that for adults. Sub-Saharan Afrca had an unemployment rate of 7.7%. The unemployment rate for women was 8.7% whle for men was 6.9%. South Afrca s faced wth the challenge of an ncreasng unemployment rate. Accordng to Statstcs South Afrca (2014), South Afrca has an unemployment rate of 25.2% (5.1 mllon unemployed persons). The unemployment rate for women s consstently above that of men. Women had an unemployment rate of 27.0% whch s 3.3 percentage ponts hgher than that of men (23.7%). Of the 5.1 mllon unemployed persons, 3.4 mllon persons were youth, whle, 1.7 mllon persons were adults. 2.3.2 Tme-related underemployment In South Afrca, the ncdence of underemployed has ncreased by 62 000, from 514 000 n 2011 to 602 000 n 2014. Women experenced hgher ncdence of underemployment compared to men. Women accounted for 5.4% whle men experenced only 2.9%. The hghest ncdence was observed amongst black Afrcans (4.7%) whle the lowest was observed amongst whte populaton (0.9%) (Statstcs South Afrca, 2014). 6

2.3.1.3 Earnngs Accordng to Statstcs South Afrca (2014), South Afrca s medan monthly earnngs ncreased by R133 from R2 900 n 2010 to R3 033 n 2014. However, the country s stll faced wth hgh ncome dstrbuton nequaltes. Men and women had medan monthly earnngs of R3 500 and R2 600 respectvely. Whte populaton had R10 000 medan monthly earnngs compared to R3 083 for coloured populaton; and R2 800 medan earnngs for black Afrcan populaton. 2.4 STUDIES ON FACTORS THAT INFLUENCE UNEMPLOYMENT Several studes on factors assocated wth unemployment have been conducted n the developed and developng natons across the world. The sectons below focus on revewng prevous studes conducted on unemployment. 2.4.1 Gender A study on gender transton between unemployment and employment was conducted n three Eastern European countres (Hungary, Poland and Slovaka). The study ncluded persons aged 21 years and older n the labour force (employed and unemployed). Logstc regresson analyss was used and t was found that gender s sgnfcantly assocated wth unemployment. In addton, the odds that men are unemployed were 1.5 tmes that of women n Hungary (Fodor, 1997). Marks & Flemng (1998) nvestgated factors nfluencng youth unemployment n Australa. The study utlsed panel data of four cohorts of Australan young people born between 1961 and 1975 from the Australan Young People Survey of 1980-1994. Logstc regresson was appled on the data to better descrbe the phenomena. The results ndcated that the odds that men are unemployed were 1.5 tmes that of women. Hazans, et al. (2003) studed factors assocated wth unemployment n Estona, Latvan, and Lthuanan. The sample of ths study ncluded jobseekers aged 15 to 55 years. Logt regresson method was used to analyse the Labour Force Survey (LFS) 1999-2000 data. It was establshed that men had hgher unemployment rate compared to women n all these countres seven percentage ponts hgher n Estona; and two percentage ponts hgher n both Latva and Lthuanan respectvely. 7

2.4.2 Age In ther study, Marks & Flemng (1998) the on factors nfluencng youth unemployment n Australa. The panel data of four cohorts of Australan young people born between 1961 and 1975 from the Australan Young People Survey of 1980-1994 was used. Logstc regresson was appled on the data to better descrbe the phenomena. They found that age s sgnfcantly assocated wth unemployment. Furthermore, ths research revealed that for one addtonal ncrease n age, the odds of beng unemployed ncreases by between 0.8 and 0.9 percentage pont. Grogan & Van den Berg (2001) compared unemployment outcomes between young people (29 years and below) and adults (30 to 60 years). A longtudnal householdbased survey data collected between 1994 and 1996 n Russa have been used to descrbe the dynamcs of unemployment on age. Cox proportonal hazard model was utlzed to establsh that ndvduals younger than 29 years are more lkely to be unemployed compared to those aged 30 years and above. An unemployment study on people wth traumatc bran njury aged 19 to 70 years n South Carolna, Unted States of Amerca, was conducted by Pckelsmer, et al. (2003). South Carolna hosptal follow-up regstry 1999 to 2002 data was used to perform the analyss. Logstc regresson was used to conclude that the odds that persons aged 45 to 70 years are unemployed were 2.14 tmes that of persons aged 45 years and below. Mahlwele (2008) conducted a study on factors assocated wth unemployment n South Afrca. Ths study ncluded women n the workng-age populaton (15 to 65 years). The Labour Force Survey (LFS) 2007 data collected by Statstcs South Afrca was utlzed. Logstc regresson analyss results ndcated that the odds that women aged 45 to 64 years are unemployed were 8.293 tmes the odds of persons aged 15 to 29 years. 2.4.3 Populaton group A research commssoned by Wu & Eamon (2011) focusng on the patterns and correlates of nvoluntary unemployment and underemployment n sngle mother famles whch utlsed 2004 panel data of the Survey of Income and Program 8

Partcpaton(SIPP) n the Unted States of Amerca. Logstc regresson analyss method was used to analyse the data. The results showed that populaton group s sgnfcantly assocated wth unemployment. The results concur wth Pckelsmer, et al. (2003) who found that populaton group s sgnfcantly assocated wth unemployment. In addton, t was establshed that the odds of beng unemployed as a non-whte were 1.60 tmes that of whte populaton. Department of Labour (1993) used Current Populaton Survey 1993 data collected on households ndvdual aged 16 years and above. The results showed that the unemployment rate for blacks were between two to 2.5 tmes that for whte populaton. 2.4.4 Martal status Several studes on the assocaton of unemployment and martal status were conducted. In ther study, Pckelsmer, et al. (2003) found that the odds that sngle persons are unemployed are 1.47 tmes that of marred persons. Hazans, et al. (2003) found that beng dvorced or wdowed ncreases the odds of beng unemployed compared to those that are marred. The study results by Foley (1997) and Stetsenko (2003) support ths fndng. Kupets (2006) studed the determnants of unemployment duraton n Ukrane usng the Ukranan Longtudnal Montorng Survey (ULMS) 1998 2002 data. The sample used ntervewed households members aged 15 to 72 years who were ether employed or unemployed. Cox proportonal hazard model was used to analyse the data. It was found that there s a strong relatonshp between martal status and unemployment. Furthermore, the odds of beng unemployed for non-marred persons were 5.714 tmes that for marred persons. 2.4.5 Level of educaton Mncer (1991) studed the relatonshp between educaton and unemployment amongst famles n the Unted States of Amerca. The target populaton was whte males aged 18 to 60 years from a survey conducted by the Survey Research Centre at the Unversty of Mchgan (1977 and 1978). Logstc regresson analyss was appled on the data. It was found that the odds of those wth less than secondary 9

educaton that are unemployed were 2.7 tmes that for those wth tertary educaton. In addton, t was found that one addtonal year n schoolng reduces the odds of unemployment by 1.3. Kngdon & Knght (2000) conducted a study on the ncdence of unemployment n South Afrca. Two datasets were used to carry the analyss namely ntegrated household survey 1993 conducted by the South Afrcan Labour Research Unt(SALDRU) and the October Household Survey 1994 conducted by Statstcs South Afrca. Probt analyss was used to conclude that havng hgher educatonal status reduces the odds of beng unemployed. McKenna (1996) found that workers wth hgher levels of educaton had two or three tmes lower chances of beng unemployed compared to those wth lower levels of educaton. Tertary educaton reduces unemployment rate by 8 to 10 percentage ponts compared to those wth secondary educaton and below (Hazans, et al. (2003)). Begum (2004) nvestgated the effect of educaton on unemployment n the Unted Kngdom usng 2003 labour market data. It was concluded that the odds that persons wth no qualfcatons are unemployed were three tmes that for persons wth tertary qualfcatons. Kupets (2006) found that persons wth tertary educaton have 0.171 chance of beng unemployed compared to those wth secondary and lower levels of educaton. A study on the effect of educaton on unemployment n ten European countres was conducted. Natonal Labour Force Survey 1975 to 2002 data whch ncluded persons aged 15 to 64 years was utlsed. Ordnary Least Squares (OLS) method was used to carry out the analyss. It was found that a percentage pont ncrease n the share of educated people reduces the unemployment rate by 0.5 percentage pont (Bag & Lucfora, 2008). Altbeker & Storme (2013) nvestgated the level of unemployment rate wthn graduates n South Afrca. The October Household Survey (OHS) 1995-1999, Labour Force Survey (LFS) 2000-2005, and Quarterly Labour Force Survey (QLFS) 2005-2011 datasets collected by Statstcs South Afrca were used. The analyss was 10

restrcted to those aged 15 to 65 years. Persons wth secondary educaton and no schoolng were 2 to 3 tmes more lkely to be unemployed compared to those wth tertary educaton. 2.4.6 Geographc locaton Grogan & Van den Berg (2001) compared unemployment outcomes and geographcal locaton n Russa. A longtudnal household-based survey data collected between 1994 and 1996 have been used analyses the dynamcs of unemployment on geographcal locaton. Cox proportonal hazard model was utlzed to conclude that those ndvduals who resde n Moscow and St. Petersburg have hgher chances of beng employed than those who resde n other areas. Hazans, et al. (2003) studed factors assocated wth unemployment n Estona, Latvan, and Lthuanan. The sample of ths study ncluded jobseekers aged 15 to 55 years. Logt regresson statstcal method was used to analyse the Labour Force Survey (LFS) 1999-2000 data. They establshed that resdng n the rural areas ncreases the odds of beng unemployed compared to resdng n the urban areas. Kupets (2006) studed the assocaton between unemployment and geographc locaton n Ukrane usng the Ukranan Longtudnal Montorng Survey (ULMS) 1998 2002 data. The sample used households members aged 15 to 72 years who were ether employed or unemployed. Cox proportonal hazard model was used to analyse the data. It was establshed that persons resdng n urban areas are 0.158 less lkely to be unemployed compared to those resdng n rural areas. 2.4.7 Concluson Ths secton revewed studes on factors assocated wth unemployment conducted by other researchers. In these studes, the followng varables were found to be sgnfcantly assocated wth unemployment: gender, age, populaton group, martal status, level of educaton, and geographc locaton. 11

2.5 STUDIES ON FACTORS THAT INFLUENCE TIME-RELATED UNDEREMPLOYMENT Ths secton revews prevous studes done on factors assocated wth tme-related underemployment by other researchers across the world. 2.5.1 Gender A study on the factors assocated wth underemployment n the rural areas of Inda was conducted by Paul (1988). The Natonal Sample Survey 1977, 1978 and 1983 data collected on employed persons. Intervewed Indvduals were classfed as ether unemployed or underemployed. The underemployment rate for women were 6.78% compared to the 3.16% for men. Thus, women s underemployment rate was twce that for men. Ellot (2004) nvestgated the lnk between underemployment and gender n the metropoltan areas of Unted States of Amerca. Ellot appled logstc regresson on the Publc Use Mcro-data Seres (PUMS) 1990 data wth the target populaton (aged 18 to 64 years). It was establshed that the odds of beng underemployed for women were about 1.15 tmes that for men. The 1998 1999 Ghanaan Lvng Standards Survey (LSS) data was used to study the nfluence of gender on underemployment. The analyss was restrcted to persons between the age of 16 and 50 years. Logstc regresson was appled and the results show that the odds of women beng underemployed were 1.036 tmes that for men (Sackey & Ose, 2006). Kjeldstad & Nymoen (2010) used the 2005 Norwegan Labour Force Survey data to nvestgate the relatonshp between underemployment and gender. The sample ncluded persons between the age of 20 and 66 years. Logstc regresson model was appled on the data and concluded that the odds of women beng underemployed are four tmes that for men. The mpact of demographc varables such as gender on underemployment was studed n Australa. The underemployment Workers Survey 2009 data where persons aged 15 years and above were ntervewed n ther sampled households. The survey 12

results showed that women are more lkely to be underemployed compared to men 9.3% compared to 5.4% (Australan Bureau Statstcs, 2010). 2.5.2 Age Lchter (1988) used the 1970-1982 Current Populaton Survey data to nvestgate the effect of age on underemployment n the central ctes of the Unted States of Amerca. The target populaton of the study were men aged 18 to 64 years who were grouped as ether unemployed or underemployed were ncluded n the data analyss. Logstc regresson was used to descrbe the contrbuton of age on underemployment. The results ndcated that the odds that persons aged 18 to 29 years are underemployed were 1.961 tmes that for those aged 30 to 54 years. In the prevously mentoned study by Ellot (2004), the assocaton between underemployment and age was also nvestgated. The results showed the odds of beng underemployed for persons aged 18-24 years were about 1.4 tmes that for those aged 45-64 years. Wlkns (2004) also studed the nfluence of age on underemployment usng the 2001 Household, Income and Labour Dynamcs Survey data collected by the Melbourne Insttute of Appled Economc and Socal Research n Australa. Persons aged 15-65 years who were ether unemployed or underemployed were ncluded n the analyss. Probt regresson was appled on the data to the establshment that the odds of persons aged 45-54 years are underemployed s 0.907 tmes that for those n other age groups. In ther study of Kjeldstad & Nymoen (2010), the assocaton between underemployment and age was also revewed n Norway. The Norwegan Labour Force Survey 2005 data was used to carry the analyss. Indvduals aged 20 to 66 years. Persons n the study were classfed as ether employed or underemployed. Logstc regresson method was used to analyse the data. It was establshed that the odds that persons aged 20-24 years are underemployed were 2.9 tmes that for those aged 45-54 years. 13

Young (2012) studed the characterstcs of underemployment n the Unted States of Amerca. Data used for ths study was Current Populaton Survey 2005-2012 collected by the Unversty of Mnnesota. Persons ncluded n the analyss were restrcted to those aged 18 years and above. The results showed that persons aged 18-39 are more lkely to be underemployed compared to those aged 40 years and above 21.7% compared to 12.2%. 2.5.3 Populaton group Tpps & Gordon (1985) revewed the nequaltes at work by populaton group n the Unted States of Amerca. The Current Populaton Survey (CPS) 1980 data to descrbe the underemployment rate by populaton group was used. Persons ncluded n the research were those aged 16 years and above. It was establshed that the underemployment rate for whte populaton has consstently been lower compared to those of other populaton groups. In the prevous mentoned study by Lchter (1988), the nfluence of populaton group on underemployment was nvestgated. It was found that the proporton of underemployment ncreased by 2.6 percentage ponts from 1970 to 1982. In addton, the odds that black Afrcans are underemployed were 1.352 tmes that for the whte populaton. Young (2012) also establshed that the odds that blacks are underemployed were 6.3 tmes those for the whte populaton. 2.5.4 Martal Status Wlkns (2004) usng the 2001 Household, Income and Labour Dynamcs Survey data collected by the Melbourne Insttute of Appled Economc and Socal Research n Australa. Persons aged 15-65 years who were coded as ether unemployed or underemployed were ncluded n the analyss. Probt regresson was appled on the data. Wlkns found that the odds of marred persons beng underemployed s 0.947 tmes that for those for those persons who were never marred. Sackey & Ose (2006) also nvestgated the nfluence of martal status on underemployment n Ghana. Data from the Ghanaan Lvng Standards 1998 1999 Survey conducted by Ghana Statstcal Servce was utlsed to carry the analyss of 14

ths research. The age of the target populaton ranged from 16 to 50 years and were grouped as ether unemployed or underemployed. Descrptve statstcs and logt model were used to carry the analyss of the research. The results revealed that the odds that marred persons are underemployed s 0.906 tmes that for those wth other martal statuses. 2.5.5 Level of educaton Underemployment s more common to those wth lower levels of educaton compared to those wth hgher levels of educaton (Australan Bureau Statstcs, 2010). Ellot (2004) nvestgated the lnk between underemployment and level of educaton n the metropoltan areas of Unted States of Amerca. Ellot appled logstc regresson on the Publc Use Mcro-data Seres (PUMS) 1990 data wth persons aged 18 to 64 years. It was found that the odds of beng underemployed for persons who dropped out of hgh school were about 1.273 tmes those for those wth tertary educaton. A research study by Wlkns (2004) on the mpact of level of educaton on underemployment usng the 2001 Household, Income and Labour Dynamcs survey data collected by the Melbourne Insttute of Appled Economc and Socal Research n Australa has been used. Persons aged 15-65 years were classfed as ether unemployed or underemployed n the analyss. Probt regresson was appled on the data to conclude that the odds for persons wth tertary educaton to be underemployed s 0.931 tmes that for those wth lower levels of educaton. In ther study, Sackey & Ose (2006) used the Ghanaan lvng standards 1998 1999 survey data conducted by Ghana Statstcal Servce. The age of the partcpants ranged from 16 to 50 years. The partcpants were grouped as ether unemployed or underemployed. Logt model was used to carry the analyss. It was found that the odds for persons wth tertary educaton beng underemployed s 0.821 tmes that for those wth other levels of educaton. Kjeldstad & Nymoen (2010) also nvestgated the relatonshp between level of educaton and underemployment. The 2005 Norwegan Labour Force Survey data was used. The target populaton of the research were persons between the age of 20 and 66 years. Logstc regresson model was appled on the data to nfer that persons wth 15

tertary educaton are 0.50 tmes less lkely to be underemployed than those wth prmary educaton. 2.5.6 Geographc locaton A study by Sackey & Ose (2006) whch used the Ghanaan lvng standards 1998 1999 survey data conducted by Ghana Statstcal Servce. The age of the partcpants ranged from 16 to 50 years. The partcpants were classfed as ether unemployed or underemployed. Logt model was used to nvestgate the nfluence of geographc locaton on underemployment. They found that the odds for persons resdng n the urban areas to be underemployed s 0.894 tmes that for those resdng n the rural areas. Fndes, et al. (2009) studed factors related to underemployment n rural Pennsylvana n the Unted States of Amerca. The Rural Pennsylvana Current Populaton Survey 1996-2006 data was used and the target populaton s age ranged from 15 to 64. The target populaton were stratfed as ether unemployed or underemployed. The study concluded that persons resdng n the metropoltan areas are less lkely to be underemployed compared to those not resdng n the metropoltan areas 20% versus 25%. 2.5.7 Sector In ther study, Sackey & Ose (2006) used the Ghanaan lvng standards 1998 1999 survey data conducted by Ghana Statstcal Servce. The age of the partcpants ranged from 16 to 50 years and they were grouped as ether unemployed or underemployed. Logt model was used to nfer that there s a strong relatonshp between underemployment and poverty amongst persons workng n the agrcultural sector. In addton, t was found that workng n the formal sector reduces the probablty of beng underemployed by 10%. Kjeldstad & Nymoen (2010) studed the assocaton between underemployment and sector. The 2005 Norwegan Labour Force Survey data was used. The study ncluded persons from the age of 20 to 66 years. Logstc regresson model was appled on the 16

data to nfer that the odds of beng underemployed for persons n the publc sector muncpal were 1.5 tmes that for those workng n the prvate sector. 2.5.8 Contract duraton In a study by Wlkns (2004) mentoned earler on, the assocaton between contract duraton and underemployment was nvestgated. It was found that part-tme or casual contract s sgnfcantly assocated wth underemployment. Tam (2010) studed the characterstcs of underemployment and the overemployed n the Unted Kngdom. The Labour Force Survey 2009-2010 data where persons age ranged from 16 years and above was used. It was establshed that the proporton of underemployment amongst full-tme workers was 5.9% compared to 20.9% for those workng part-tme. Kjeldstad & Nymoen (2010) concluded that the odds for persons wth temporary contract are underemployed were 3.5 tmes that for those wth a permanent contract. 2.5.9 Concluson Prevous studes on factors assocated wth tme-related underemployment were revewed. These studes showed that gender, age, populaton group, martal status, level of educaton, geographc locaton, sector, and contract duraton were sgnfcantly assocated wth tme-related underemployment. 2.6 STUDIES ON FACTORS THAT INFLUENCE EARNINGS Many researchers across the world studed factors assocated wth earnngs. As such, ths secton revews varables whch were found to be sgnfcantly assocated wth person s earnngs. 2.6.1 Gender Angle & Wssman (1981) conducted a comparatve study on young employed persons. The Natonal Longtudnal Surveys of Labour Market Experence data collected between 1968 and 1975 for women, and 1966 and 1975 for men was used. In the analyss, men and women had dfference age categores where the age for men 17

ranged from 21 to 33 years whle that for women ranged from 21 to 31 years. Logstc regresson was appled on the data to conclude that gender s sgnfcantly assocated wth earnngs. In addton, the research found that the odds that women belong to hgh earnngs s 0.66 tmes that for men. A survey was conducted n Bogota, Colomba n 1988, whch collected nformaton on educaton, earnngs, employment and other personal characterstcs of workers wth an average age of 30. Lnear regresson was appled on the data to conclude that gender s sgnfcantly assocated wth earnngs. Furthermore, the results showed that men earn 30 percentage ponts hgher than women. Male graduates earn 72,30 percentage ponts hgher than female graduates (Psacharopaulos & Velez, 1992). Report by the Unted States Government Accountablty Offce (USGAO) whch studed pay dfferences between men and women n the Unted States of Amerca usng the Current Populaton Survey (CPS) 2010 data. The analyss of the report was lmted to employed persons age rangng from 25 to 64 years. Logstc regresson was used to conclude that the odds that women belong to low earnngs were 1.64 tmes that for men (USGAO, 2011). 2.6.2 Age Blnder (1973) found that age s sgnfcantly assocated wth earnngs. Can, et al. (1973) found that age has a postve and sgnfcant effect on earnngs of people aged between 45 and 49 years. Hrsch (1978) conducted a study on earnngs, occupaton, and human captal nvestment. The focus of ths study was to nfer whether age plays an mportant role n determnng persons earnngs amongst other thngs. The data used was the Census 1967-1970 conducted n the Unted States of Amerca. In the analyss, employed whte people who do not resde n the farms and are not male students aged between 15 and 64 were ncluded. Regresson analyss was appled on the data to conclude that earnngs ncreases wth age. A comparson study on earnngs between young (20-34 years) and older (45-54 years) persons n the Unted States of Amerca was conducted by Freeman (1979). The 18

Natonal Income and Product Accounts (NIPA) 1970 1983 and Current Populaton Survey (CPS) 1973 1982 datasets were used to carry the analyss. It was found that the earnngs for older men were 1.55 tmes that for younger men. In addton, the earnngs for older women were 1.15 tmes that for younger women. Wenberg (2004) nvestgated the earnngs dfferences between men and women n the Unted States of Amerca. The Census 2000 data wth the lmtaton of employed persons aged 16 years and above was utlsed n ths research. The results showed that people aged 34 years and above earn more than those aged 34 years and below. Ths results agrees wth the fndngs by the USGAO report whch found that persons aged 25-34 years had the hghest percentage of persons n the lower earnngs compared to those aged 34 years and above (USGAO, 2011). The Bureau of Labour Statstcs (2014) studed the assocaton of weekly earnngs and age n the Unted States of Amerca by usng the 2013 Current Populaton Survey data. The age of employed persons ncluded n the analyss ranged from 16 years and above. It was concluded that persons aged 55-64 years have the hghest medan earnngs compared to those aged 16 24 years - $1,048 compared to $474. 2.6.3 Populaton group The mpact of populaton group on earnngs was studed n England. Data from Cyrl Burt, and Thomas and Margaret Harrell (1962-1964) was used. The data collected ncluded whte and black men workng n the Armed Forces. Lnear regresson was used to best descrbe the assocaton between race and earnngs. The results of the research showed that whte persons earn three tmes hgher than that for blacks (Brown & Reynolds, 1975). Chswck (1980) studed earnngs of whte and coloured male mmgrants n Brtan. General Household Survey 1972 data was used to conduct analyss of employed persons aged 25 to 64 years. Logstc regresson was appled on the data to conclude that the odds that coloured men belong to hgher earnngs are 0.25 tmes that for whte men. 19

Whte populaton group accounted for about 20% of the South Afrca s populaton, however, they receved almost 70% of the country s earnngs over the perod (1917 to 1970) (Seventer, et al. (2000)) n South Afrca. Bhorat, et al. (2009) nvestgated the assocaton between populaton group and earnngs. The October Household Survey 1995 data was utlsed. They found that the probablty of beng on lower earnngs was hgher amongst blacks compared to those n other populaton groups. Ths results agrees wth the fndngs by Lang (2012) who reported that Whte South Afrcans earn sx tmes more than that of Black South Afrcans. The report by the Bureau of Labour Statstcs(2014) whch used the Current Populaton Survey 2013 data to better understand the assocaton of populaton group and earnngs found that black men and women earn 72.1% and 85.3% compared to whte men and women earnngs respectvely. 2.6.4 Martal status The 1968 and 1974 Swedsh Level of Lvng Survey data were used to study the wage rates and personal characterstcs n Sweden. Data analyss ncluded employed men n pad work. Lnear regresson analyss was used to best descrbe personal charactersts assocated wth earnngs. It was found that martal status s sgnfcantly assocated wth earnngs. The results further revealed that marred men have hgher average wage rates than wdowers and never marred men (Blomqust, 1979). Korenman & Neumark (1990) studed the relatonshp between martal status and earnngs usng Natonal Longtudnal Survey of Young Men (NLSYM) collected, between 1976 and 1980, by the center for Human Resources Research n the Unted States manufacturng frm personnel fle. The target populaton was whte men who completed schoolng n 1976 where the youngest was 24 years old. The results showed that marred men earn more than men wth other marred status. In addton, never marred men have lower wages than men n other martal status group. In ther research, Blomqust (1979) and Korenman & Neumark (1990) fndngs agreed wth the results of USGAO (2011) that marred persons earn more than never marred persons. 20

Data from the 1979 and 1986 Luxembourg Income Study (LIS) were used to best explan the dfferences on eanngs by martal status. The data contaned nformaton on measures of ncome and socal well-beng of persons n the developed countres. The sample was restrcted to those men who are head of households and are between 25 and 55 years. Lnear regresson was appled on the data to conclude that martal status s sgnfcantly assocated wth earnngs. Futhermore, marred men earn 30 percentage ponts more than those who are not marred; separated men earn 15-25 percentage ponts more than those who have never marred (Scheon,1995). 2.6.5 Level of educaton Educaton plays a key role n provdng ndvduals wth knowledge, sklls and competences needed to partcpate effectvely n socety and n the economy. Lfetme earnngs ncreases wth each level of educaton attaned (OECD,2012 & Julan,2012). Hrsch (1978) and Blomqust (1979) nvestgated the effect of educaton on earnngs. They found that educaton plays an mportant role n determnng ndvduals earnngs. Chswck (1980) conducted a study on the arnngs of whte and coloured male mmgrants n Brtan. General Household Survey 1972 data was used to conduct analyss of employed persons aged 25 to 64 years. Logstc regresson was used to conclude that an extra year of schoolng s assocated wth 7.3% earnngs ncreases. Accordng to Psacharopaulos & Velez (1992) who used a survey conducted n Bogota, Colomba n 1988, whch collected nformaton on educaton, earnngs, employment and other personal characterstcs of the workers wth an average age of 30 years to perform the analyss. Lnear regresson was used to nfer that hgher levels of educaton s strongly assocated wth hgher levels of earnngs. Ashraf & Ashraf (1998) nvestgated the mpact level of educaton on eanngs. The Soco-Economc Survey of Karach, 1987 1988, was conducted by the Appled Economcs Reseach Centre at the Unvesty of Karach was used n the research. The age of persons ncluded ranged from 15 years and above. Regresson analyss results showed that the odds for persons wth masters qualfcaton and above belong to hgh earnngs were 3.25 tmes that for those n other levels of educaton. 21

Accordng to the Robert Wood Johnson Foundaton (2009) hgher levels of educaton attaned s assocated wth hgher-payng employment. Meanwhle, women wth tertary educaton earn three tmes that for those women wth hgh school educaton. In addton, t was also establshed that one addtonal year of schoolng represents 11% ncrease n earnngs. Wu (2011) used logstc regresson to revew factors assocated wth long term employment and low-ncome on mothers. It was found that the odds that women wth more than hgh school educaton belongs to hgher earnngs were three tme that for those women wth hgh school educaton. 2.6.6 Geographc locaton A study concernng earnngs nequalty n Brazl was commssoned by Lam & Levson (1992). Logstc regresson was appled on the Current Populaton Survey 1985 data to descrbe the stuaton. The analyss of the research ncluded employed men. The results showed that the odds for persons resdng n the urban areas belongs to hgher earnngs were four tmes that for those resdng n the rural areas. Bhorat, et al. (2009) used the 1995 October Household Survey Data and Labour Force Survey data to study the mpact of geographc locaton on earnngs n South Afrca. Logstc regresson was used to revew the mpact of geographc locaton on earnngs. The study found that persons resdng n the urban areas earn 13 percentage ponts hgher than those resdng n the rural areas. Wu (2011) used logstc regresson to revew factors assocated wth long term employment. It was found that locaton was sgnfcantly assocated wth earnngs. The study results further showed that the odds for persons resdng n the east belong to hgher earnngs were 2.27 tmes that for persons resdng n the west regon. 2.6.7 Unon membershp Vencarachellum & Mchaud (2001) nvestgated the assocaton of unon membershp and earnngs amongst black populaton n South Afrca usng Lvng Standards and Development (LSD) 1993 data. The unts of analyss were employed men and women. It was establshed that men and women who are afflated to labour unons earn 26 22