Are jobless persons who want work but are not actively searching, unemployed or out of the labour force? Previous research on this issue has focused on North America and used as the test whether the probability of transition to employment is similar for searching and non-searching jobless persons. The research program is developing new tests as to whether those not searching but wanting work are distinct from the searching unemployed. It asks: are non-searching persons richer, happier, and do they have a lesser impact on local wages, than the searchers? These tests are implemented using data from South Africa, a country in which unemployment is very high and where the treatment of non-searchers really matters for the understanding of poverty and labour market issues. The results favour the ‘discouraged worker' view of the non-searching unemployed and the use of the broad, inclusive, measure of unemployment.

This work is using three new approaches to test the distinctness of the searching and non-searching unemployed states in conditions of high unemployment. We find, firstly, that in South Africa the non-searching unemployed are, on average, significantly more deprived than the searching (see Table 1 below). The fact that they are not better-off casts doubt on the interpretation based on tastes (lack of desire for employment) and favours the interpretation that active search is discouraged (low prospective returns to search). This view is supported by the evidence from a job-search model, which suggests that search is hampered by poverty, cost of job-search from remote rural areas, and high local unemployment. Secondly, the non-searching unemployed are not significantly happier than the searching unemployed. Their unemployment depresses their wellbeing to the same extent as is the case for the searching unemployed. This result proved to be robust to various checks. Finally, evidence on the wage-unemployment relationship indicates that local wage determination takes non-searching workers into account as genuine labour force participants. The searching and the non-searching unemployed are very close in terms of potential labour supply.
 

Table 1

Sample of Unemployed Persons: Type of Unemployment and Socio-economic Outcomes, SALDRU93 data

	Non-searching unemployed	Searching unemployed	t-statistic for the test of the null hypothesis that the two means are equal
Unemployment rate
- Cluster unemployment rate	0.493	0.417	-12.01
- Household unemployment rate	0.763	0.727	-4.45
Per capita household income :
(Rand/month) - mean	162.541	232.234	5.37
(Rand/month) - median	96.673	133.125
Per capita household expenditure:
(Rand/month) - mean	190.887	282.079	8.35
(Rand/month) - median	136.017	184.069
Other indicators:
Below international poverty line $1/d	0.485	0.381	-6.48
Number of assets	2.931	3.619	9.28
Years of education	6.793	7.595	6.75
African	0.953	0.796	-13.53
Household size	7.228	6.554	-5.56
Living conditions:
Lives in a house/part of house	0.473	0.561	5.38
Number of household members per room	1.965	1.906	-1.41
Home is owned	0.773	0.705	-4.61
Dwelling has corrugated iron roof	0.681	0.571	-6.92
Piped water within or tap in yard	0.363	0.571	12.88
Has to fetch water daily	0.603	0.395	-12.91
Distance to water (meters)	305.931	170.513	-7.14
Dwelling has flush toilet	0.269	0.461	12.11
Dwelling has electricity connection	0.302	0.451	9.33
Community characteristics:
Urban	0.352	0.599	15.39
Homeland	0.667	0.431	-14.69
Distance to facilities from home	111.096	74.124	-8.79
Community has tarred roads	0.100	0.236	10.65
Community roads impassable at certain times of year	0.563	0.402	-9.91
N	2775	1379
Percentage of the total labour force	20.8%	10.4%

Notes : Apart from ‘years of education', all the non-community variables above are coded at the household level in the dataset. For the purposes of this table, however, we have assigned the value of the household variable to each individual member of the household, then taken the sub-sample of unemployed persons only (2775 non-searching and 1379 searching unemployed persons) and averaged the variables across individuals. Similarly, the community variables are assigned to each individual living in that community before averaging across unemployed individuals. The very high household unemployment rate indicates that unemployed people are likely to live in households where other members are unemployed as well. Of the 13316 labour force participants aged 16-64 in this dataset, 4154 were unemployed.

The appropriate concept of unemployment necessarily depends on the purposes to which the ensuing measure will be put. In South Africa unemployment is an important determinant of poverty and of wellbeing. A measure of unemployment is required which can illuminate anti-poverty policies. Two of the tests under this research are appropriate for this use of the unemployment measure. Unemployment also has an influence on wages and other labour market behaviour. Our third test is relevant in selecting the measure appropriate for gauging the state of the labour market. Taken together, the three tests suggest that there is no distinction between those searching and those not searching that would warrant the exclusion of the non-searching from the measure of unemployment or any less policy concern for the non-searching than for the searching unemployed.

The evidence from South Africa is consistent with the prediction that very high unemployment will cause many of the unemployed to be discouraged workers. In such conditions, the non-searching unemployed are no less part of the labour force than the searching unemployed and their joblessness is no less associated with deprivation.

Better data, such as a nationwide panel, are ideally required to be more certain that the relationships found in this research are causal rather than the result of simultaneity or unobserved heterogeneity. Cross-section findings cannot be definitive and, on this issue and many others, Statistics South Africa needs to move towards longitudinal data gathering in its (annual) October Household Surveys. For instance, the analysis of transition probabilities would be possible with panel data. However, the test that has been variously employed – whether the probability of transition to employment is higher for the searching than the non-searching unemployed - might fail to convince in conditions of high unemployment. If the non-searchers were deterred from active search by low job prospects and high search costs – actions which in turn further reduced their probability of employment – that would not necessarily mean that they wanted work any less than the searchers.

Given the sheer magnitude of unemployment and the 12-15 percentage point gap in the broad and narrow unemployment rates, the appropriate definition and measure of unemployment is an extremely important question for South Africa which deserves the best possible answer using the data that are currently available. It might be thought that unemployment, being so high, is sure to receive priority in policy-making even if the narrow definition is used. However, it is arguable that the political alliance between the ruling party (ANC) and the trade union organisation (Cosatu) encourages labour market policies which are inimical to employment creation and to unemployment reduction. The effective playing down of the unemployment problem through the use of the narrow measure lightens this counterweight to pro-union policies.

Clearly, there is an internationally agreed ILO definition of unemployment, and it is very helpful for international comparisons if each country calculates it. The research here does not propose that the broad measure replace the ILO-recommended narrow measure, even in South Africa. However, our findings imply that the non-searching unemployed deserve no less policy attention than do the searchers, and that the broad unemployment measure should be estimated alongside the narrow measure and given credence, in South Africa and in other high unemployment economies.