[ Pobierz całość w formacie PDF ]
.The survival function is calculated asS(t) = t (1 - qj).j=1The hazard of job loss we report is calculated in the usual way using the quarteras the time unit,ht = qt / (1- qt /2). 120 King and MueserThis approximation is based on the assumption that the hazard of job loss isconstant throughout quarter t.For t > 1, this is a reasonable approximation in ourdata, but it is not correct for t = 1 because jobs do not generally begin at the startof the quarter.We discuss this issue below.4.Missing data in quarters 1995:3 and 1997:2 in Fort Lauderdale required specialattention.Survival and hazard calculations omit any spell beginning in a quarterfor which data are missing as well as spells beginning in the following quarter,since it is not possible to determine whether employment first observed in thefollowing quarter actually began in that quarter.We used the ratio for the spelllength distribution observed for spells with available data to determine the num-ber of spells terminating during the two quarters hidden by the missing data.We also used an additional correction to account for the possibility that a spellspanning a quarter with missing data actually discontinued prior to the missingquarter but that the individual began a new job spell with the same employer im-mediately after the missing quarter, based on experiments in which we omitted aquarter of actual data.5.We have chosen not to adjust the first-quarter hazard rate for the expected dateof job start because there are clearly other serious inaccuracies that are relevantin the first period.It is known that the hazard of job loss increases after the firstfew weeks on a job, followed by a dramatic decline.Since the biases are similaracross sites, there is no difficulty in comparing first-period hazards.6.For these analyses, only job spells that could last a full eight quarters can beincluded, and so spells used in the analysis for the period 1998 1999 differ bysite according to data availability.For Baltimore, Fort Lauderdale, and KansasCity, only spells beginning in the first quarter of 1998 are included.For Chicago,spells beginning in the first three quarters of 1998 are included, and for Atlanta,spells beginning in 1998 and the first quarter of 1999 are included.No informa-tion on spells beginning in Houston in 1998 or 1999 is available.Houston datafor 1996 1997 include only information on spells beginning through the firstquarter of 1997.7.All reported earnings have been adjusted to real dollars for the fourth quarter of1999 based on the U.S.Consumer Price Index for urban consumers.As in theprevious analyses, in Houston, information on job spells as reported in Tables 4.1and 4.2 extends only through the first quarter of 1997.8.There is no government-sanctioned measure comparable to the Consumer PriceIndex that attempts to capture cost of living differences across cities in the Unit-ed States.One commercially produced index suggests that, among our sites, costof living is highest in Fort Lauderdale, nearly 30 percent above the U.S.aver-age, with Chicago about 20 percent above the national average, and Atlanta 10percent above the national average.Our other sites are 5 to 10 percent belowthe national average (http://houseandhome.msn.com).These cost-of-living es-timates bear little relation to observed differences in earnings reported in Table4.1.It should be recognized that not only are such measures crude, but they arenot designed to reflect the costs faced by welfare recipients.9.As above, special calculations were necessary to obtain earnings measures in Job Stability for Welfare Recipients 121Fort Lauderdale because of missing data in two quarters.We omit any spell be-ginning in a missing quarter or any spell beginning in the following quarter.For job spells beginning in other quarters, we constructed the spell omitting themissing quarters of data, allowing a job to span across the missing quarter.For allanalyses in which second quarter earnings were matched, we omitted any spellwith a second quarter corresponding to the quarter with missing data.For analy-ses reported in Tables 4.1 4.3, which require measures of earnings for each job,we substituted data from other spells in Fort Lauderdale as follows: For a spellmissing earnings data in a particular quarter, we selected spells beginning withinthe same general time period (1992 1993, 1994 1995, or 1996 1997) that wereof the same length as the spell in question and with the same welfare status, fill-ing in the data from a random spell.Where we were unsure of the length of thespell because the spell ended immediately prior to a missing quarter, we matchedthe spell randomly with a spell from a population containing spells of both pos-sible lengths, taking from the matching spell both the spell length, and, whennecessary, the missing earnings information.10.Chapter 5 examines the impact of individual and employer characteristics on jobstability for welfare recipients.11.Employers are defined according to the account maintained in the state unem-ployment insurance program.In general, all employees of a firm within a stateare defined as being employed by a single employer, while employees outside thestate are excluded.12
[ Pobierz całość w formacie PDF ]