UNEMPLOYMENT BENEFIT LEVELS AND SEARCH ACTIVITY

OXFORD BULLETINof

ECONOMICS and STATISTICSVolume 55 February 1993 No. 1OXFORD BULLETIN OF ECONOMICS AND STATISTICS. 55. 1(1993)0305-9049 $3.00

UNEMPLOYMENT BENEFIT LEVELS ANDSEARCH ACTIVITY

John Schmitt and Jonathan Wadsworth *

I. INTRODUCTION

In standard search models of unemployment, the probability that anunemployed worker enters a job equals the probability that the workerreceives a job offer times the probability that the job is acceptable to theworker. An acceptable job pays a wage that exceeds the worker's 'reservationwage', the wage where a worker is indifferent between working and remainingunemployed.

Most theoretical analyses of the effect of state benefits on unemploymenthave focused on the way that government assistance raises workers' reserva-tion wages. These models assume that job offers either arive at some constantrate or by some exogenously-determined stochastic process. Workers revieweach offer and reject those which are below their cut-off wage. While it isdifficult to establish a worker's reservation wage, most empirical research hasconcentrated on demonstrating the notion expounded by these early searchtheories (e.g. Mortensen (1970)) that higher benefit levels prolong theduration of unemployment spells by reducing the expected return fromemployment.

Given the strength of the theoretical predictions concerning benefits andreservation wages, the evidence on duration and transition effects of benefits

* We are grateful to seminar participants at the LSE, ANU, EALE, and UCL, and also toSteve Nickell and an anonymous referee for helpful comments. Material from the GeneralHousehold Survey, made available through the Office of Population Censuses and Surveys andthe ESRC Data Archive, has been used by permission of the Controller of HM StationeryOffice. The Centre for Economic Performance is financed by the Economic and SocialResearch Council.

1

2 BULLETIN

is not strong. In Britain, for example, estimated benefit duration elasticitiesrange from 0.06 to 0.6.' In the United States, with a different unemploymentcompensation system, many estimates of benefit duration elasticities rangearound 0.4 (see Atkinson and Micklewright (1990) for a comprehensivereview). A notable outlying study by Clark and Summers (1983), however,finds that in a large panel of male workers, higher unemployment benefitlevels were associated with a higher probability of moving into employment.2For the Netherlands, van den Berg (1990) finds a very low, positive elasticityof unemployment duration with respect to changes in the benefit level(approximately 0.03).

Recent evidence suggests that the unemployed receive and reject very fewjob offers. Using data on a cross-section of unemployed British workers in1982, Jones (1989a) reports that over 85 percent have never received ajoboffer. Holzer (1988) presents a figure of 66 percent in similar results for maleyouths in the United States in 1981. Van den Berg (1990), using a panel ofunemployed Dutch men, finds low job offer arrival rates (approximately aone percent chance per week) and a very high proportion of acceptable offers(approximately 97 percent for the whole sample). Survey evidence reportedin Jackman et aL (1991) suggests that unemployed men in Britain make a low,but regular, number of job applications - around one or two a month -which tend to diminish over the unemployment spell.

The experience of many workers therefore seems to be that they remainunemployed primarily because they receive few job offers, not because theyreject many. If so, then the simple reservation wage effect of benefits is opento question. Within the context of search models, it is conceivable thatunemployment benefits may still affect job matching by influencing the rate ofjob offer arrival. This seems particularly pertinent, since recent work findingthat receipt of unemployment benefits increases search extensiveness amongthe unemployed (see Blau and Robins (1990) for the United States andWadsworth (1991) for the United Kingdom), raises the possibility thatexpanding benefit coverage could increase the search effectiveness of thepool of unemployed.

Unemployment-related benefits might alter search extensiveness in twoways. First, unemployed benefit recipients may use their greater resources tofinance search activity, other things equal. For example, the number of appli-cations a worker makes may depend positively on the level of incomesupport. Conversely, search activity could decline if the relative returns toemployment fall as unemployment benefits rise. Second, receipt of statebenefit may increase ties to the labour force through information or incentive

'On this basis, given an average uncompleted duration of 12 months and an estimatedelasticity of 0.3, then a 10 percent increase in benefit levels, for example, would raise averagespell duration by around 11 days.

2 effects were positive but not different from zero at standard levels of significance.

UNEMPLOYMENT BENEFIT LEVELS AND SEARCH ACTIVITY 3

effects.3 For example, benefit payments may be contingent on recipientsdemonstrating active job search. Those not claiming state benefit maybecome marginalized in their attempts to locate work.

This paper is a formal test of the first hypothesis. If benefits encouragerecipients to devote more (or less) resources to search activities, then the levelof benefit itself could affect the degree of search effort. We utilize informationfrom the General Household Survey to examine the job search behaviour of asample of unemployed male benefit recipients in Great Britain over theperiod 1 979-82. We are therefore able to test whether job search activity, asmeasured by the number of search methods used in a given week, responds todifferential levels of government assistance. Section II provides a brief outlineof the underlying theoretical framework. Section III describes the data andthe principal estimation procedure, ordered logit maximum likelihood.Section IV presents results on the choice of search method and extensivenessof search effort, which are ambiguous with regard to the direction of anybenefit effect. Section V concludes with a brief summary of our majorfindings that job search activity is largely uninfluenced by the level ofunemployment compensation but is dependent on worker characteristics andenvironment.

II. THEORETICAL FRAMEWORK

Standard search theory posits that the probability an unemployed workerenters employment, Pe, is given by

P=O*A (1)

where O is the probability that a job offer is received and A is the probabilitythat it is acceptable. Most models assume that the rate of job offer arrival isexogenous. If, however, workers can influence the job offer arrival ratethrough search activity, then O becomes endogenous such that

O=O[mS(B,g,x)] (2)

The function, S(.) defines the 'search effort' of the unemployed individual.The more diligently the workers search, the more likely they are to find a job.The individual's search effort depends on several factors: the level of un-employment benefit, B, local economic conditions, g, and a vector ofpersonal characteristics, x, which includes the potential wage from employ-ment. In addition, m is a multiplicative constant which reflects the state of thebusiness cycle.4

third possible explanation for more extensive search is unobserved heterogeneitybetween claimants and non-claimants. Wadsworth (1991) attempts to control for this selectivitybias with respect to labour force attachment.

m represents the total number of jobs available in any time period. Therefore, job offersmay arrive at differential rates for a given level of search effort.

4 BULLETIN

Since A also depends on the benefit level, equation (3) demonstrates thatbenefits could have an ambiguous effect on the job transition probability,dependent on the signs of the two constituent effects.

dP , 3S AO (3)

We proceed to examine the effect of benefit levels on search effort, aS/aB, asa first step toward addressing this potential ambiguity.

Standard models predict higher levels of benefit should decrease searcheffort since the relative return to search falls as unemployment benefit rises(see for example, Barron and Mellow (1979)). On this basis, one might expectto see the number of job applications decline, or the number of offer refusalsincrease, as unemployment compensation rises. Given our dataset, thishypothesis suggests a fall in the number of search methods used at any onetime. Alternatively, a rise in benefit may have positive effects on search. Ifworkers use this income to finance more search activity this may offset anyincreased preference for leisure. This may arise, for example, if unemployedworkers are credit-constrained (see Hammermesh (1982)) or leisure is alocally inferior good, as suggested by the results in van den Berg (1990). Ben-Horim and Zuckerman (1987) demonstrate the possibility of such a positiveeffect, in a model where the unemployed finance search from limitedresources. Given these two opposing effects, it the sign of any benefit leveleffect on overall search effort becomes a matter for empirical investigation.

LII. DATA AND ESTIMATION

The data consist of a pooled sample of observations on 1,484 unemployedBritish men participating in the General Household Survey (GHS) between1979 and 1982. The GIS is a representative annual survey of approximately13,000 British households, which seeks to ascertain detailed information onhousehold composition and activities.5 Four GHS surveys, from 1979 to1982, provide consistent data on the type of search methods used by un-employed respondents in the week prior to their GHS interview. In addition,the GHS contains self-reported levels of state benefits received by eachrespondent (with a breakdown by source), and the total state benefit andother income received by all members of their families.

The GHS asks unemployed workers whether they have used any of six jobsearch methods in the week prior to their interview. The methods are: visiteda Job Centre, used a private employment agency, read or placed an advertise-nient in the newspaper, directly approached an employer, waited for theresults of a job application, or some other method of job search.6 Responses

utilize the full version of the OHS which oversaniples Scottish households.'The omission of use of personal contacts and the amalgamation of reading and placement

of advertisements among the choice of search methods are unfortunate aspects of the data.


to these questions create two kinds of dependent variables. The first is asimple O-1 dummy for the use of a particular method. The second is a searchextensiveness variable calculated as the total number of search methods usedin the reference week of the survey.

The central problem facing any attempt to estimate the effects of benefitson search effort is the latter's inherent unobservability. Many factors affectsearch activity: the amount of resources devoted to search effort; the number(and 'quality') of hours spent searching; the willingness to compromise ongeographical location, working conditions, and pay; the type, number, andappropriateness of search methods used; differential individual qualitieswhich affect search effectiveness; and others.

Empirical work has generally taken one of the list of components above asa proxy for search effort.7 The number of search methods says little abouthow intensively the individual uses each method. The choice of number ofsearch methods then, could lead us to conclude that a lazy searcher choosingseveral methods in a half-hearted way is exerting more 'effort' than a diligentindividual using just one method intensively.8

We believe, nevertheless, that the number of search methods is a goodproxy for underlying search effort. Wadsworth (199 la) demonstrates a smallpositive relationship between the number of search method used and theprobability that a worker finds a job. Furthermore, the effects of variables forwhich we have strong prior beliefs with respect to search effort - unemploy-ment duration, educational qualifications, marital status, and age - can act aschecks for the benefit results. We show in Section 1V that these variablesalways take their expected signs in the number of search method regressions.

All unemployed workers in the dataset receive either UnemploymentBenefit or Supplementary Benefit, the two state benefit programmes pro-viding, respectively, short-term and long-term assistance to the unemployed.The majority of the sample unemployed for less than seven months wouldalso have been eligible for Earnings Related Supplement (ERS) to theirunemployment benefit during the period of our analysis.9 We use the sum ofUnemployment Benefit and Supplementary Benefit (a small portion ofrespondents claim both) to calculate the respondent's nominal level ofbenefits, converted to real January 1979 pounds using the Retail Price Index

7For example, Jackman and Williams (1985) use the number of job applications, Jones(1989) utilizes hours of search, and Blau and Robins (1990) have data on the number ofcontacts made by unemployed workers.

Note that the total number of hours spent searching, a very common proxy for search effort,can perform poorly if it excludes information about the total number of search methods. Forexample, a worker may use one search method for a period of ten hours while another may usetwo methods over the same ten-hour period. An hours-based proxy would assign the twoindividuals the same search effort. However, if search methods exhibit diminishing returns orcomplementarities, the second individual will have exerted more 'effort'.

9ERS was discontinued in January 1982, but existing claimants were eligible until June ofthat year. The dataset does not give any indication as to the specific amount of ERS receivedout of total benefit.

6 BULLETIN

appropriate for the interview month and year. This levels variable, RBEN,can be thought of as capturing the income effect of benefits on search. Inorder to measure the benefit effect on the return to search it is necessary totake account of expected future benefit levels relative to potential wageopportunities. This we do by constructing, REPRAT, a 12-month forwardmoving average of expected real benefit levels deflated by an expected wageterm obtained from a simple earnings equation (see data appendix for furtherdetails).

The GHS also asks respondents about alternative sources of income,earned, unearned, and from other state benefits. We can therefore constructvariables which measure the family's total earned income and total receipt ofall fonns of state benefit (excluding the unemployed worker's unemploy-ment-related benefits). We have supplemented the data by attaching severalexternal labour market variables to workers based on their region, skill level,and the year of their interview with the GHS. We used these variables(vacancy, unemployment and wage rates) to control and test for businesscycle and expected return effects on search behaviour, as suggested byequation (2). Our preferred measure is the relevant vacancy-unemploymentratio.

From the standpoint of econometric estimation, the structure of the benefitsystem (see Appendix 2 for details) and the pooled nature of the dataset,allow several types of variation in the level of benefits across otherwiseeconometrically identical workers: (1) policy related changes ¡n basic andsupplemental rates (including eligibility requirements), for UB and SB; (2)changes in real UB and SB rates due to monthly changes in the price level; (3)variations in the ERS component for the portion of the sample unemployedbetween 2 and 28 weeks. Figure 1 demonstrates this variation over thesample period, together with the smoothed expected benefit series, for atypical sample member - a single male with no dependent children°

Estimation

Ordered logistic analysis is ideally suited for estimating the sign of EiS/aB in(3). We have observations on a discrete, ordered variable, the number ofsearch methods, y, which reflect the distribution of an unobservable, under-lying variable of interest, search effort, ya". Given that 46.2 percent of oursample use more than one search method, then estimation by ordered logitgives us more information about the effect of benefits on search than would asimple binary logit which combined the number of methods used into aunitary measure. The latent variable, y, is influenced by a vector of variables,x, incorporating benefits, socio-economic characteristics and local labour

10 The rise in the expected benefit series in the last year of the sample, reflects the real rise inbenefit levels that occurred in 1983. Removal of the final year from our sample does not affectour results.

17.5

17-

16.5 -

16 -

15.5

15-

14.5-

14-


Real Unemployment BenefitSingle Male with No Children

13.5

1979 1980 1981 1982

Notes: (1) Source: HM Treasury (1985) and authors' calculations.The jagged line shows the nominal unemployment benefit level for a single malewith no dependent children, deflated by the monthly retail price index. Thenominal rate rises in December of each year.The smoother line shows the 'expected' real benefit for the same individual basedon the procedure used to create the REPRAT variable. (See the data appendix.)

Fig. 1

market conditions, such that:1'

y* = ß'x + e

If there are J ordered number of search methods, we observe:

y0 jf *<Çj

y=l if

y=2 if tS1<U2

y=J if /4j_y*

"The variables in the x vector in the log-likelihood equation appear in the Appendix 1.They include personal and family characteristics, proxies for external labour market conditions,and measures of financial resources available to the individual.

8 BULLETIN

where the threshold levels j. (j = 1, 2, ... , J-1), are parameters of the modelto be estimated along with ß (see Greene (1990), Maddala (1983)). We definea set of ordinal variables:

Z,1 = 1 if y falls in the jth category, and

Z - O if ydoes not,

j = 1, 2, ... , n individuals, j = 0, 1, ... , J search methods.

If e is logistically distributed and A represents its cumulative distribution,then:

Prob(Z, = 1) 'A(u1 ß'x)A(u1_1 ß'x)

We can write the likelihood function as:

L = fl fl [A(M1ß')A(._1 - ß'x)ji1 j-1

And the log likelihood function is:

L* = in L= Z,1ln[A( ß'x1)A(1 ß'x)ji=1 j-1

We maximize the log-likelihood function using the iterative techniques avail-able in UMDEP.

Interpreting Coefficient Estimates

The coefficient estimates for the variables in an ordered logit model do nothave a straightforward interpretation. In general, a positive coefficient esti-mate suggests that 'high' values of the variable mean an individual is likely touse more search methods. However, to evaluate the marginal effects of achange in the value of a particular variable note that the probability that ytakes a particular value is:

Prob[y= 01= A(ß'x)

Prob[y= 1]=A(,u1ß'x)A(ß'x)Prob[y= 2]A(,u2ß'x)A(,u1 ß'x)

Prob[yJ]1A(.t_1 ß'x)


The marginal effect of a change in a variable in x is then:

8Prob[y= O]v(ß'x)ß

8Prob[y= 1(v(-ß'x)- v( -ß'x))ßax

aProb[ y = J]V(/4j_1 -ß'x)ß

ax

where y is the logistic density function.

IV. RESULTS

Choice of Search Method

Table 1 presents a breakdown of search methods used according to socio-economic group, uncompleted duration of unemployment spell, and yearinterviewed. Approximately 85 percent of the total sample used one or moresearch methods in the week prior to their interview. The average number ofsearch methods used is 1.6. Job Centres were the most important searchmethod, with over 70 percent visiting a Job Centre in the previous week. Thispartially reflects the nature of the UK benefit administration scheme inoperation between 1979 and 1982, whereby unemployed claimants were

TABLE lASearch Method Use by Occupation

Sample SOCI SOC2 SOC3 SOC4 S005 Entrant

None 16.6 16.9 12.1 17.0 16.9 17.8 16.7Job Centre 69.6 51.9 71.0 71.3 70.3 70.2 64.1Private 5.5 24.7 12.1 3.8 3.8 3.1 3.8Advert 19.9 29.9 29.8 19.8 16.9 17.3 17.9Direct 27.1 28.6 21.0 30.8 26.7 24.6 16.7Wait 24.0 37.7 33.1 25.9 18.2 18.8 28.2Other 17.0 31.2 17.7 17.0 18.0 11.0 11.5

MedianNSM 1 2 2 2 1 1 1

Mean NSM 1.6 2.0 1.8 1.7 1.5 1.5 1.4Total NSM 2431 157 229 970 687 277 111N 1484 77 122 571 445 191 78

10 BULLETIN

TABLE lBSearch Method Use by Unemployment Duration

Duration (months)

<1 l-3 3-6 6-12 >12

None 6.8 12.7 10.5 16.2 22.9Job Centre 76.7 71.0 74.8 72.2 64.0Private 6.8 8.1 8.5 4.5 3.3Advert 27.4 27.1 21.2 23.0 13.8Direct 42.5 35.3 30.7 25.6 21.0Wait 42.5 32.1 28.1 23.9 16.5Other 27.4 24.4 19.9 16.2 11.9

MedianNSM 2 2 2 1 1

Mean NSM 2.2 2.0 1.8 1.7 1.3TotaINSM 163 438 561 511 758N 72 220 305 308 579

TABLE 1CSearch Method Use by Year

Y79 Y80 Y81 Y82

None 19.2 14.6 16.5 17.4Job Centre 62.3 74.2 71.5 65.5Private 5.4 3.4 6.1 6.8Advert 21.3 22.6 19.0 17.0Direct 31.8 31.8 24.6 22.7Wait 29.3 25.8 21.6 22.7Other 18.0 18.3 15.7 17.8

Median NSM 1 2 1 1

Mean NSM 1.7 1.8 1.6 1.5TotalNSM 402 615 1011 403N 237 347 637 263

also registered at Job Centres. About one-third of the sample had made adirect approach to a potential employer. Hughes and McCormick (1990)suggest that this activity is likely to reflect greater intensity of purpose leadingto shorter unemployment durations. However, use of this method is likely tobe highly pro-cyclical. It seems reasonable to suppose that the unemployedadapt their search strategies to maximize the possibility of locating a job offer.Hence we would not expect direct approaches to employers to featurestrongly when unemployment is high and increasing as during the period


covered by our sample. Approximately 20 percent were waiting for aresponse from a job application, using advertisements, or trying othermethods. Only 6 percent had sought work through a private employmentagency.

The worker's occupational classification (SOC) has a strong influence onthe choice of search method. The GHS assigns occupational classifications tounemployed workers based on the skill-level of their last job. Non-manualsare approximately five times more likely than manuals to use a private agencyand are more likely than manuals to use advertisements, job applications('Waiting'), and 'other' techniques. Manuals, particularly skilled-manuals, aremore likely to use Job Centres and direct approaches to employers. Table 1also makes clear that differential use of search techniques appear within themanual and non-manual categories.'2

The mean number of search methods declines monotonically with uncom-pleted duration of unemployment spell. The number of workers using nosearch method in the previous week increases threefold between the firstmonth of unemployment and the period after the first 12 months. As dura-tion lengthens search efforts appear to concentrate around Job Centres. Thebreakdown of search activity by year reveals no systematic pattern in aggre-gate job search behaviour over the sample period, despite the large rise inunemployment which occurred between 1979 and 1982.

Table 2 illustrates how the unemployed combine search methods in theirattempts to locate job offers. The correlation coefficients in Table 2a confirmthe complementarity of most search methods. Table 2b indicates the popu-larity of Job Centres as the most favoured search tactic, followed by directapproaches to potential employers. Use of the media accompanies mostother forms of job search.

Table 3 reports binary logistic estimates of the determinants of individualsearch method use. The results confirm those of the cross-tabulations inTable 1. Choice of search method is organized broadly across occupationallines. Those groups which theory suggests should have a low probability ofsearch effort (the long-term unemployed, the old, the unqualified), areindeed less likely to use any individual search method. Further evidence thatthe aggregation of each method used constitutes an appropriate measure ofsearch effectiveness.

Benefits have no significant effect on the choice of any individual searchmethod. The benefit coefficients in the no methods used regression take thehypothesized signs - in this case negative for the income effect of RBEN,positive for the relative return effect of REPRAT - but are far from sig-nificant.

The level of economic activity, as measured by the appropriate VU ratio, ispositive and significant for job search through newspaper advertisements.

12Separate evidence from the GHS on successful job search methods used by 861 recentlyhired workers in 1980, provides additional support for the notion that job type is a majordeterminant of the choice of search methods.

12 BULLETIN

TABLE 2ACorrelation Matrix of Job Search Method Use

** Significant at 95% level.

TABLE 2BSearch Method Complementarities

Job Centre Private Advert Direct Other Wait

Job Centre 1

Private 0.017 1

Advert 0.107** 0.186** 1

Direct 0.107** 0.093** 0.318** 1

Other 0.004 0.025 0.026 - 0.036Wait 0.083** 0.198** 0.348** 0.245** 0.030

Sample numbers using each method

Tight labour markets increase the worker's perception of contacting avacancy in this manner. Similarly, the sign of the VU ratio on directapproaches to employers, whilst not significant, suggests use of this method isalso more common under more favourable economic conditions.'3

Financial Resources and the Number of Search Methods

Our central objective is to test the hypothesis that differential benefit levelsaffect job search effort. Table 4 presents our preferred model, estimated byordered logit. Column 1 includes the appropriate regional vacancy-unemployment rate facing each worker. Column 2 uses regional and yearlydummies to capture business cycle effects. Table 6 presents marginal effectsof key variables using the results from Table 4, column 1.

and McCormick (1990) find employer contact search increases significantly withimprovements in local economic conditions.

Job Centre Private Advert Direct Other Wait

Job CentrePrivateAdvertDirectOtherWait

103359

233313178272

8141361748

293163

56158

40360

166254

68 356


The estimated model includes four variables which measure the financialresources available to each unemployed worker: RBEN, RSTBEN, RINCand REPRAT. RBEN, the real level of unemployment-related benefit, andREPRAT, the expected replacement ratio facing each worker, are the mainvariables of interest. Our preferred specification in column 1 indicates asmall, statistically insignificant positive effect of benefit levels on search effortand a similar negative effect of the replacement ratio.'4 The derivatives of thecategory probabilities with respect to the level of RBEN and REPRAT (seeTable 6) confirm that differential levels of unemployment-related benefithave a negligible impact on the likelihood of using more search methods. Apermanent 20 percent increase in the average level of benefits would reducethe likelihood of an average worker using more than one search method byless than one-half of a percent (Table 6, Row 1).

Table 5 presents several alternative specifications intended to test therobustness of the result. Columns 1, 2 and 3 split the sample according toduration of unemployment spell. It is possible that RBEN and REPRAT maycapture unobserved heterogeneity amongst recipients. If those previouslyemployed in high wage jobs, who would therefore qualify for high ERS, aremore 'keen' to return to employment, they may search harder. This may biasupward both the RBEN and REPRAT coefficients. The division of thesample by duration allows us to control for this effect somewhat, since onlythose unemployed for under seven months were eligible to receive ERS. Thesign and significance of the RBEN coefficient in column 1 suggests littlesubstantive bias. Hence, the decomposition of the sample into what are essen-tially UB + ERS and SB recipients does not affect the conclusions.'5

Column 4 excludes all those who have used no search method during thereference week, in order to test for benefit effects amongst a more 'active'subset of job seekers. Again there is little change in the benefit coefficient.

Column 5 addresses potential endogeneity in the model specification.Since those choosing fewer search methods could face longer unemploymentspells, other things equal, the inclusion of duration variables may induceendogeneity bias in the benefit variable. We therefore excluded duration as anexplanatory variable in column 5. Again the results are substantiallyunchanged. Since current search activity should explain potential rather thanelapsed duration, endogeneity may be less of an issue here.'6

the record, estimation with only the income variables on the right-hand side gives thefollowing results:

.029*Rben.017*Reprat LogL= 2286.1(.007) (.005) X2(4)= 49.57

'51t may be argued that the UK benefit administration system which, until 1982, meantbenefit claimants were closely linked with local job centres could bias the benefit variable.However, since the entire sample consists of claimants, who were presumably subject to thesame regime, this should not influence the results.

' See also Lindeboom and Theeuwes (1991), who manipulate the joint probability ofobserving a given level of search and unemployment duration to show that search intensity canbe estimated as a function of lapsed unemployment duration.

TA

BL

E 3

Log

istic

Est

imat

es f

or C

hoic

e of

Job

Sea

rch

Met

hods

Var

iabl

eM

ean

Non

eJo

b C

entr

ePr

ivat

eA

dver

tD

irec

tW

ait

Oth

er

CO

NST

AN

T1.

00-2

.499

1.14

3-2

.139

-1.6

83-0

.645

-1.0

66-1

.474

(0.3

88)*

*(0

.302

)**

(0.6

58)*

*(0

35Ø

)**

(0.3

11)*

*(0

.334

)**

(0.3

70)*

*

RB

EN

100

24.6

7-0

.012

-0.0

110.

033

-0.0

020.

007

0.00

60.

002

(0.0

28)

(0.0

22)

(0.0

46)

(0.0

24)

(0.0

22)

(0.0

24)

(0.0

26)

RST

B10

09.

120.

004

0.00

70.

004

-0.0

02-0

.004

-0.0

05-0

.014

(0.0

06)

(0.0

05)

(0.0

11)

(0.0

06)

(0.0

05)

(0.0

06)

(0.0

07)*

*

RIN

C10

045

.30

0.00

10.

001

0.00

1-0

.001

0.00

10.

003

0.00

1(0

.002

)(0

.011

)(0

.003

)(0

.011

)(0

.001

)(0

.001

)**

(0.0

02)

RE

PRA

T37

.76

0.01

7-0

.002

-0.0

200.

007

-0.0

11-0

.010

-0.0

13(0

.019

)(0

.014

)(0

.003

)(0

.016

)(0

.015

)(0

.016

)(0

.018

)

VU

R0.

08-0

.040

-0.3

78-0

.107

1.82

00,

666

0.74

40.

476

(0.9

18)

(0.7

16)

(1.4

67)

(0.7

87)*

*(0

.747

)(0

.788

)(0

.846

)

DU

RM

T0.

210.

475

-0.1

26-0

.589

-0.0

19-0

.397

-0.2

23-0

.300

(0.2

16)*

*(0

.166

)(0

.350

)*(0

.179

)(0

.167

)**

(0.1

74)

(0.1

94)

DU

RL

T0.

390.

800

-0.4

32-0

.740

-0.5

00-0

.587

-0.3

94-0

.547

(0.1

88)*

*(0

.144

)**

(0.3

25)*

*(0

.174

)**

(0.1

51)*

*(0

.162

)**

(0.1

81)*

*

A16

190.

13-0

.523

0.17

0-0

.709

-0.0

41-0

.141

0.23

20.

388

(0.5

38)

(0.4

11)

(0.9

23)

(0.4

64)

(0.4

21)

(0.4

43)

(0.4

94)

A20

240.

18-0

.510

0.48

8-0

.578

-0.0

540.

263

-0.1

070.

393

(Ø.3

Ø4)

*(0

.234

)**

(0.4

78)

(0.2

61)

(0.2

29)

(0.2

49)

(0.2

76)

A25

340.

22-0

.601

0.11

6-0

.599

0.26

60.

250

0.20

90.

608

(0.2

51)*

*(0

.178

)(0

.372

)(0

.197

)(0

.176

)(0

.187

)(0

,210

)**

A50

640,

240.

892

-0.3

57-0

.674

-0.7

07-0

.795

-1.0

32-0

.404

(0.2

13)*

*(0

.178

)**

(0.3

83)*

(0.2

33)*

*(0

.204

)**

(0.2

23)*

*(0

.243

)*

Not

e: A

sym

ptot

ic t-

stat

istic

s in

par

enth

eses

. Sam

ple

size

1484

.**

indi

cate

s si

gnif

ican

ce a

t the

95

perc

ent l

evel

. * in

dica

tes

sign

ific

ance

at t

he 9

0 pe

rcen

t lev

el.

MA

RR

iED

0.50

-0.2

060.

402

-0.1

420.

168

0.24

10.

461

0.10

7(0

.221

)(O

.181

)**

(0.3

91)

(0.2

18)

(0.1

93)

(0.2

07)*

*(0

.226

)E

TH

NIC

0.06

0.00

10.

072

-0.2

090.

208

0.52

2-0

.102

-0.3

74(0

.311

)(0

.249

)(0

.577

)(0

.272

)(0

.241

)**

(0.2

72)

(0.3

36)

DC

HIL

D0.

40-0

.133

0.11

9-0

.182

-0.0

03-0

.011

0.04

8-0

.046

(0.2

16)

(0.1

70)

(0.3

71)

(0.1

96)

(0.1

72)

(0.1

82)

(0.2

04)

CO

UN

CIL

0.58

0.03

60.

090

-0.8

19-0

.253

0.17

9-0

.403

-0.0

32(0

.181

)(0

.143

)(0

.300

)**

(0.1

63)

(0.1

49)

(0.1

53)*

*(0

.172

)PR

iVA

TE

0.12

0.36

1-0

.110

-0.4

72-0

.444

-0.1

40-0

.214

-0.3

51(0

.243

)(0

.199

)(0

.411

)(0

.252

)*(0

.226

)(0

.227

)(0

.269

)z

ED

120.

04-0

.963

-0.2

102.

162

1.57

30.

439

0.83

2-0

.035

(0.5

82)*

(0.3

29)

(0.4

09)*

*(0

.338

)**

(0.3

39)

(0.3

36)*

*(0

.389

)E

D34

0.11

0.39

2-0

.421

1.00

00.

309

-0.3

180.

463

-0.0

50(0

.274

)(0

.211

)**

(0.4

09)*

*(0

.233

)(0

.221

)(0

.217

)**

(0.2

51)

ED

50.

19-0

.027

-0.1

290.

717

0.39

80.

053

0.45

90.

309

(0.2

02)

(0.1

55)

(0.3

46)*

*(0

.181

)**

(0.1

66)

(0.1

68)*

*(0

.180

)*

SOC

10.

050.

170

-0.6

580.

665

0.30

00.

439

0.50

30.

741

(0.4

03)

(0.3

07)*

*(0

.532

)(0

.354

)(0

.339

)(0

.339

)(0

.356

)**

ZSO

C2

0.08

-0.2

870.

063

0.35

50.

477

-0.3

130.

543

0.01

2(0

.319

)(0

.234

)(0

.412

)(0

.250

)*(0

.258

)(0

.241

)**

(0.2

79)

SOC

30.

390.

150

0.00

3-0

.436

-0.0

260.

132

0.17

1-0

.096

(0.1

86)

(0.1

49)

(0.3

52)

(0.1

75)

(0.1

52)

(0.1

66)

(0.1

82)

EN

TR

AN

T0.

050.

319

-0.4

96-0

.547

-0.1

21-0

.638

0.21

3-0

.407

(0.3

88)

(0.2

96)*

(0.7

50)

(0.3

63)

(0.3

52)*

(0.3

21)

(0.4

14)

Log

L-6

14.0

3-8

77.7

9-2

57.1

3-6

88.9

5-8

24.6

4-7

47.6

9-6

44.8

0L

ogO

-666

.48

-911

.38

-314

.29

-737

.30

-867

.85

-817

.62

-679

.26

Mea

nD.V

.0.

166

0.69

60.

055

0.19

90.

271

0.24

0.17

16 BULLETIN

TABLE 4Ordered Logistic Estimates: Number of Search Methods

Independent variable Sample means (1) (2)

CONSTANT 1.00 2.172** 2.089**(0.257) (0.295)

RBEN 24.67 0.012 0.002(0.0 18) (0.020)

RSTB 9.13 -0.003 -0.001(0.004) (0.004)

RINC 54.42 0.00 1* 0.00 1(0.001) (0.001)

REPRAT 37.76 -0.013 -0.006(0.012) (0.014)

VUR 0.081 0.704(0.606) -

DURMT 0.207 - 0.336** - 0.342**(0.132) (0.133)

DURLT 0.390 _0.776** _0.750**(0.120) (0.122)

A1619 0.132 0.231 0.095(0.345) (0.373)

A2024 0.178 0.291 0.228(0.187) (0.192)

A2534 0.222 0.358** 0.370**(0.147) (0.148)

A5064 0.241 _0.892** _0.913**(0.150) (0.151)

MARRIED 0.497 0.285* Ø349**

(0.150) (0.152)

ETHNIC 0.064 0.143 0.019(0.191) (0.196)

DCHILD 0.396 0.065 0.098(0.137) (0.142)

COUNCIL 0.582 -0.155 -0.155(0.118) (0.120)

PRIVATE 0.121 _Ø345** _0.376**(0.162) (0.164)

ED12 0.040 1.127** 1.157**(0.26 1) (0.263)


TABLE 4- contd

Note: (1) Standard errors in parentheses.(2) ** indicates significance at 5% level; * indicates significance at 10% level.

Column 6 removes the income effect to leave a standard search-benefitspecification which includes REPRAT alone. The coefficient is negative, ashypothesized, but remains small and statistically insignificant (t = 1.48).

We therefore find no evidence to support standard search theory that thelevel of benefits are the channel through which unemployment programmeschange search effort. Rather, it seems likely that the positive coefficient found

Independent variable Sample means (1) (2)

ED34 0.106 0.013 0.037(0.169) (0.173)

ED5 0.195 0.205 0.205(0.130) (0.134)

SOC1 0.052 0.249 0.351(0.245) (0.252)

SOC2 0.082 0.238 0.254(0.192) (0.194)

SOC3 0.385 -0.005 0.057(0.122) (0.126)

ENTRANT 0.053 -0.375 -0.352(0.270) (0.274)

MU(1) - 1.962** 1.986**(0.076) (0.077)

MU(2) - 3.184** 3.220**(0.092) (0.094)

MU(3) - 4.315** 4359**

(0.117) (0.119)

MU(4) - 6.008** 6.058**(0.213) (0.216)

Dummies:Regions (9) - No YesYears (3) No Yes

LogL - -2171.0 -2158.3LogO - -2286.1 -2286.1D.E - 23 34

N - 1484 1484

Note: All regressions include the explanatory variables in Table 4, Column 1.

by Blau and Robins (1990) and Wadsworth (1991) of benefit receipt onnumber of search methods, is due to informational and incentive effects ofsuch programmes which lead claimants to search more extensively relative tonon-claimants.

RINC, the real total income available to the worker's family (less all formsof state benefit) does, however, exhibit a significantly positive effect on thetotal number of search methods. This result provides some support for thenotion that financial constraints limit the ability of the unemployed to engagein active job search. Since RINC rises with employed household members, itmay capture better job information flows available from these parties.However, the magnitude of these effects are small.'7

'7The coefficient on RINC may be smaller than for other forms of income since this incomeaccrues to other family members. The unemployed worker may not have the same claims overthese resources as over unemployment-related benefits. The removal of RINC from Table 4does not affect the sign or significance of RBEN.

18 BULLETIN

TABLE 5Alternative Search Activity Specifications

Independent (1) (2) (3) (4) (5) (6)variabk STU MTU LTLJ NSM>O

RBEN -0.009 -0.018 0.035 0.013 0.017(0.029) (0.035) (0.032) (0.020) (0.018)

RSTB -0.002 -0.002 -0.007 -0.003 -0.005(0.007) (0.010) (0.007) (0.005) (0.004)

RINC 0.002* 0.001 -0.001 ØØØ3** 0.0035*(0.001) (0.002) (0.002) (0.001) (0.001)

REPRAT 0.001 -0.004 -0.021 -0.010 -0.017 -0.007(0.019) (0.024) (0.022) (0.014) (0.012) (0.004)

DURMT - - -0.196 _Ø395**(0.148) (0.128)

DURLT - - _0.654** - 0.833(0.134) (0.115)

VUR 2.101** 1.427 -0.681 1.008 1.030* 0.861(0.885) (1.487) (1.126) (0.685) (0.595) (0.579)

LogL -915.8 -455.1 -764.6 -1542.5 -2191.8 -2173.2LogO -952.8 -470.1 -812.7 -1619.7 -2286.1 -2286.1

D.F. 21 21 21 23 21 20

N 597 308 579 1238 1484 1484


TABLE 6Change in the Predicted Probability of Using j Search Methods

Note: Figures show deviation from predicted probabilities evaluated at sample means of allindependent variables. Coefficients taken from model in Table 4, Column 1.

The insignificant negative coefficient on RSTBEN probably reflects thefact that other forms of state benefit are generally earmarked for particularpurposes such as dependent children, disabilities, illnesses, etc. As such,these other forms of state benefit do not necessarily represent net additionsto the individual's resources available for job search or consumption.'8

The results on marginal effects presented in Table 6 also support thenotion that the number of search methods is a good proxy for latent searcheffort or an unobservable exit probability. Those groups we would expect toexert greater effort - married, 'prime-age' workers, the short-term un-employed, those with educational qualifications - are less likely to fall incategories O and 1 and more likely to fall in the higher categories, than aneconometrically average worker. Many of these variables capture individual-specific differences in the valuation of employment.19

'8Combining the three income variables into a single variable measuring all potential avail-able income while unemployed yields a positive but statistically insignificant income effect onsearch.

a discussion of the problem of unobservables in binary models see Yatchew andGriliches (1985). The age and occupation variables partially capture the likelihood of theworker being laid off from the previous job, an important omission from our dataset.

AP(y=0) AP(y=l) P(y=2) ¿.P(y=3) z.P(y4) AP(y5)

RBEN, 0.004 0.005 -0.003 -0.004 -0.002 -0.001REPRAT(+ 20%)

VUR -0.007 -0.008 0.005 0.006 0.004 0.001(+1 s.d.)

DURLT 0.094 0.102 -0.064 -0.073 -0.046 -0.012SOC1 -0.025 -0.033 0.016 0.023 0.015 0.004

A5064 0.151 0.098 -0.101 -0.085 -0.049 -0.013ED12 -0.086 -0.168 0.026 0.107 0.093 0.029

EstimatedProb(y=j)at mean (ß'x)

0.135 0.391 0.264 0.131 0.064 0.015

Sample 0.165 0.371 0.239 0.133 0.072 0.018Prob(y=j)

Sampleoutcomes

246 551 355 197 107 28

20 BULLETIN

The long-term unemployed are between 9 and 11 percentage points morelikely to fall in one of the low search categories, (y=O) or (y=1). Thisuniform decline in search effort as duration rises may be attributable toduration dependence or heterogeneity in the pool of the unemployed. Thevariable with the largest (negative) impact on search effort is age (50 to 64years old). For example, workers in this age group are ten percentage pointsless likely to use two methOds than an average worker in the 'prime-age'category. These effects are large relative to the measured benefit effects.

The data provide some evidence that an improvement in local economicconditions induces increased search activity (see Table 4, column 1). This isparticularly evident when we split the sample into the short-, medium- andlong-term unemployed (Table 5, columns 1, 2 and 3). Local economic con-ditions have a significant impact on the search effort of the short-termunemployed alone. We believe this provides further evidence that the long-term unemployed become increasingly marginalized from the labour force.Moreover, this differential behaviour raises concerns that an economicrecovery may do little to reduce the ranks of the long-term unemployed.20

Nevertheless, the magnitude of the search response to local economicconditions for the whole sample is small, even among the short-termunemployed. An increase of one standard deviation (in this case a doublingfrom the sample mean of 0.085) in the average VUR, raises the probabilitythat the typical worker will use two or more methods by less than one per-centage point.

V. CONCLUSION

We set out to test the hypothesis that unemployment benefits could influencesearch effort in a model of endogenous job offer arrival. We find little supportfor the idea that benefits affect search extensiveness in either direction. Highbenefit levels can facilitate search by providing income with which to financejob search efforts. There may also be a contemporaneous disincentive effectfrom higher benefits reducing the relative return to employment. Our resultsindicate that whilst these factors may be at work, their magnitude and statis-tical significance are negligible. Hence, the level of benefit appears to exertlittle influence on overall search activity. Other potential sources of incomedo have a positive effect on search effort, though possibly as a proxy forinformation flows.

We conclude that previously observed high levels of search effort amongstbenefit recipients relative to non-recipients probably reflect participatoryfactors rather than pecuniary incentives. Receipt of benefit promotes both

and McCormick (1990) are similarly pessimistic that economic expansion willraise employment of nonsearching' official claimants. They find that overall search rates don'trespond significantly to changes in the local labour market; direct employer contact, by whatthey term active searchers', however, does follow swings in economic activity. The durationeffect we identify here may shed light on their results.


greater information flows and stronger search incentives through contact withthe benefit system.

Our results indicate that the most productive ways to increase the searcheffectiveness within the unemployed claimant pool, may be to improveeducational attainment, to limit the incidence of long-term unemployment,and to address the particular problems facing older unemployed workers.

Centre for Economic PerformanceLondon School of Economics and Political Science

Date of Receipt of Final Manuscript: March 1992

REFERENCES

Atkinson, A. and Micklewright, J. (1990). 'Unemployment compensation and labourmarket transitions: a critical review', London School of Economics, TaxationIncentives and Distribution of Income, Discussion Paper No. 143.

Barron, J. and Mellow, W (1979). 'Search effort in the labour market', Journal ofHuman Resources, Vol. 14, pp. 389-404.

Ben-Horim, M. and Zuckerman, D. (1987). 'The effect of unemployment insuranceon unemployment duration', Journal of Labour Economics, Vol. 5, No. 3, pp.386-90.

Blau, D. and Robins, P. (1990). 'Job search outcomes for the employed and un-employed', Journal of Political Economy, Vol. 98, pp. 637-55.

Clark, K. and Summers, L. (1983). 'Unemployment insurance and labor markettransitions', in M. Baily (ed.), Workers, Jobs and Inflation, Brookings Institution,pp.279-316.

Greene, W. (1990). Econometric Analysis, Macmillan.H.M. Treasury (1985). 'Tax Benefit Reference Manual', mimeo.Hammermesh, D. (1982). 'Social Insurance and Consumption; An Empirical

Enquiry', American Economic Review, Vol. 72, No. 1, pp. 101-13.Holzer, H. (1988). 'Search method use by unemployed youth', Journal of Labour

Economics, Vol.6, pp. 1-20.Hughes, G. and McCormick, B. (1990). 'Measuring unemployment and cylical par-

ticipation in the British labour market', Scandinavian Journal of Economics, Vol.92, pp. 247-69.

Jackman, R. and Williams, C. (1985). 'Job applications by unemployed men', LondonSchool of Economics, Centre for Labour Economics Working Paper No. 792.

Jackman, R., Layard, R. and Nickell, S. (1991). 'Job search: the duration of un-employment', Chapter 5 of Unemployment, Macroeconomic Performance and theLabour Market, OUP.

Jones, S. (1 989a). 'Job search methods, intensity and effects', Oxford Bulletin of Eco-nomics and Statistics, Vol. 51, pp. 277-96.

Jones, S. (1989b). 'Alternative implications of a structural model of job search',London School of Economics, Centre for Labour Economics Discussion PaperNo. 363.

22 BULLETIN

Lmdeboom, M. and Theeuwes, J. (1991). 'Search, Benefits and Entitlements',Leiden University Working Paper, June 1991.

Maddala, G. (1983). Limited Dependent and Qualitative Variables in Econometrics,Cambridge University Press.

Mortensen, D. (1970). 'Job search, the duration of unemployment and the Philipscurve', American Economic Review, Vol. 60, pp. 847-62.

van den Berg, G. (1990). 'Search behaviour, transitions to non-participation and theduration of unemployment', Economic Journal, Vol. 100, pp. 842-6 5.

Wadsworth, J. (1991). 'Unemployment benefits and search effort in the UK labourmarket', Economica, Vol. 58, pp. 17-34.

Wadsworth, J. (1991a). 'Unemployment benefits and labour market transitions inBritain', Centre for Economic Performance Working Paper No. 125.

Yatchew, A. and Griliches, Z. (1985). 'Specification error in probit models', Review ofEconomics and Statistics, Vol. 66, No. 1, pp. 134-9.

APPENDIX 1

Data Description

RBEN Total unemployment and supplementary benefit received inprevious week, in constant January 1979 pounds.

RSTB Total amount of all forms of state benefit received by allfamily members in previous week, less RBEN, in constant1979 pounds.

RINC Total amount of all forms of earned income and state bene-fits received by all family members in previous week, lessRBEN and RSTBEN, in constant 1979 pounds.

REPRAT Expected replacement ratio. Calculated as weighted averageof nominal benefit level, NOMBEN, and NOMBEN plusnext year's average percentage increase, i.e.

12m* NOMBEN+ * (NOMBEN+X%)

where m is the month of observation relative to November,the month for annual update of benefits. This is thendeflated by a 12-month forward moving average of RetailPrice Index, divided by predicted real weekly earningsbased on earnings equations estimated from 1979 to 1982GHS.

VUR Ratio of reported vacancies to unemployed workers inworker's region in the year interviewed. [Source: Depart-ment of Employment, Gazette. Vacancies refer to thoseexcluding Community Programmes and Careers Offices.]

DURMT Dummy variable equal to one if the worker has been con-tinuously unemployed between six and twelve months.


DURLT Dummy variable equal to one if the worker has been con-tinuously unemployed more than one year.

A1619, A2024, Dummy variables equal to one if the worker's age isA2534, A5064 between 16 and 19, 20 and 24, 25 and 34, and 50 and 64,

respectively.OTHEMP Dummy variable equal to one if the any member of the

worker's family is employed.MARRIED Dummy variable equal to one if the worker is married.ETHNIC Dummy variable equal to one if the worker is not white.DCHILD Dummy variable equal to one if the worker has a dependent

child.COUNCIL, Dummy variables equal to one if the worker rents aPRIVATE council flat, or rents private accommodation, respectively

(the default category for housing tenure dummies is owner-occupied housing).

ED12 Dummy variable equal to one if the worker has a universitydegree or other post A-level qualification.

ED34 Dummy variable equal to one if the worker has A-levels ormore than five 0-levels.

ED5 Dummy variable equal to one if the worker has at least one0-level or equivalent (the default category for the educationdummies is workers with no educational qualifications).

SOC1 Dummy variable equal to one if the worker's last job wasprofessional or managerial.

SOC2 Dummy variable equal to one if the worker's last job was asa junior non-manual.

SOC3 Dummy variable equal to one if the worker's last job was asa skilled manual worker (the default category for the skill-level dummies is semi-skilled, SOC4, and unskilled manualworkers, SOCS).

EN lEANT Dummy variable equal to one if the worker is a new or re-entrant to the labour force with no data on previous job.

R2-R10 Dummy variables for each of the ten standard Britishregions (excluding Northern Ireland) (the default categoryfor the regional dummies is the South East).

Y80, Y81, Y82 Dummy variables for the year in which the worker wasinterviewed (the default category is 1979).

APPENDIX 2

The Benefit System

Two programmes offered financial assistance to unemployed workers duringthe period covered in the data set. Unemployment Benefit (UB) providedassistance for the first 52 weeks of an unemployment spell. Thereafter, un-

24 BULLETIN

employed workers received assistance through Supplementary Benefit (SB)with no time limit on eligibility.

UB, first established in 1948, is a national insurance, contributory benefit.To qualify, a worker must have made national insurance contributions abovea minimum level prior to the unemployment spell. UB is not means-testedand does not depend on workers other income or assets (with a few excep-tions). The government pays UB at a fixed rate with additional supplementsfor dependents. Annual adjustments in the basic rates occurred in Novemberof each year from 1979 to 1982. During almost the entire period from 1979to 1982, UB recipients unemployed between 2 and 28 weeks could alsoqualify for an Earnings Related Supplement based on the level of theirweekly wage immediately prior to entering unemployment. For full details ofthe ERS scheme see HM Treasury (1985).

SB, introduced in 1966 to replace National Assistance, is a means-tested,non-contributory benefit. The DHSS determined the level of a recipient's SBafter assessing the household's 'resources' (earnings, other benefits, pensions,sick pay, income from capital, and other income) and 'requirements' (a basicrate based on fan-iily type and size plus housing costs, heating, and other case-specific expenses). As with UB, the government made adjustments to thebasic SB rates in November of each year in our sample.

Documents

UNEMPLOYMENT BENEFIT LEVELS AND SEARCH ACTIVITY