Analysis of Survey Data phần 8 docx

To control for endogenous selection and attrition, we keep all the random effects fixed at their median values of zero, and reset alltransition intensities into state O to zero.. Table 1

destination-specific Gamma heterogeneity uncorrelated across spells Second,the transition intensities are not generally monotonic; an increasing then fallingpattern is found for the transitions C ! YTS, E ! YTS and YTS ! U Theaggregate hazard rate is non-monotonic for exits from college, employment andunemployment Third, transition intensities for exit to college are very small forall states of origin except unemployment, where there is a sizeable intensity oftransition into education at short unemployment durations The generally lowdegree of transition into state C reflects the fact that, for most people, formalpost-16 education is a state entered as first destination after leaving school, ornot at all However, the fact that there are unobservables common to both theinitial state and transition parts of the model implies that the decision to entercollege after school is endogenous and cannot be modelled separately from thetransitions among the other three states The discontinuity of exit probabilities

of the two-year YTS limit is very marked (see Figure 16.7)

Figure 16.8 shows the aggregated hazard rates, k_(tjx, u) ˆP`k`, governingexits from each state of origin, k The typical short unemployment duration-simply a high hazard rate for exits from unemployment, but declining stronglywith duration, implying a heavy right hand tail for the distribution ofunemployment durations For the other three states of origin, the hazardrates are rather flatter, except for the one- and two-year peaks for college spells.Note that we cannot distinguish unambiguously between true duration depend-ence and the effects of non-persistent heterogeneity here, at least not withoutimposing restrictions such as proportionality of hazards

16.4.4 Simulation strategy

The model structure is sufficiently complex that it is difficult to interpret theparameter estimates directly Instead we use simple illustrative simulations tobring out the economic implications of the estimated parameter values The

1.2

E

1 0.8 0.6 0.4 0.2 0

`base case' simulations are performed for a hypothetical individual who isaverage with respect to quantitative attributes and modal with respect tomost qualitative ones An exception to this is educational attainment, which

we fix at the next-to-lowest category (GCSE2), to represent the group forwhom YTS is potentially most important Thus our representative individualhas the characteristics listed in Table 16.1

The treatment of state O (nonignorable attrition) is critical We assume that,conditional on the persistent heterogeneity terms uC, uE, uU, uYTS, the labourmarket transition process is the outcome of a set of independent competingrisks represented by the hazards k` Superimposed on this process is a fifth

4.5 4 3.5 3 2.5 2 1.5 1 0.5 0

Duration by state of origin (days)

E C

Figure 16.8 Aggregated hazard rates

Table 16.1 Attributes of illustrative individual

Educational attainment One or more GCSE passes, none above grade D

Subject mix Academic mix of school subjects

School quality Attended a school where 38.4 % of pupils achieved five or

more GCSE passesArea quality Lives in a ward where 77.9 % of homes are owner-occupiedLocal unemployment Unemployment rate in ward of residence is 10.3%

Date of episode Current episode began on 10 March 1989

Previous YTS No previous experience of YTS

Occupation When employed, is neither clerical nor craft/technicalSpecial needs Has no special training needs when in YTS

independent risk of attrition which is relevant to the observation process butirrelevant to the labour market itself In this sense, attrition is conditionallyignorable, by which we mean that it is independent of the labour markettransition process after conditioning on the persistent unobservables It is notunconditionally ignorable since the unobservables generate correlation betweenobserved attrition and labour market outcomes, which would cause bias ifignored during estimation However, conditional ignorability means that,once we have estimated the model and attrition process jointly (as we havedone above), we can nevertheless ignore attrition in any simulation that holdsfixed the values of the persistent unobservables In these circumstances, ignor-ing attrition means marginalising with respect to the attrition process Giventhe assumption of conditional independence of the competing risks, this isequivalent to deleting all hazards for exit to state O and simulating theremaining four-risk model This procedure simulates the distribution of labourmarket event histories that would, if unhampered by attrition, be observed inthe subpopulation of individuals with heterogeneity terms equal to the givenfixed values uk.

The simulation algorithm works as follows For the representative individualdefined in Table 16.1, 500 five-year work histories are generated via stochasticsimulation of the estimated model.1These are summarised by calculating theaverage proportion of time spent in each of the four states and the averagefrequency of each spell type To control for endogenous selection and attrition,

we keep all the random effects fixed at their median values of zero, and reset alltransition intensities into state O to zero We then explore the effects of thecovariates by considering a set of hypothetical individuals with slightly differ-ent characteristics from the representative individual These explore the effects

of ethnicity, educational attainment and the nature of the locality For the last

of these, we change the SCHOOL, AREA and URATE variables to values of10%, 25% and 20% respectively

Table 16.2 reveals a large impact for the variables representing ethnicity andeducational attainment, in comparison with the variables used to capture theinfluence of social background An individual identical to the base case, butfrom a non-white ethnic group (typically South Asian in practice), is predicted

to have a much higher probability of remaining in full-time education (59 % ofthe five-year period on average, compared with 20 % for the reference whiteindividual) However, for ethnic minority individuals who are not in education,the picture is gloomy Non-whites have a much higher proportion of their non-college time (22 % compared with 9 %) spent unemployed, with a roughlycomparable proportion spent in YTS

The effect of increasing educational attainment at GCSE is to increase theproportion of time spent in post-16 education from 20 % to 31 % and 66 % forthe three GCSE performance classes used in the analysis Improving GCSE

1 The simulation process involves sampling from the type I extreme value distribution for the logit parts of the model, and from the distribution of each latent duration for the transition part In both cases, the inverse of the relevant cdf was evaluated using uniform pseudo-random numbers.

Table 16.2 Simulated effects of the covariates for a hypothetical individual.

Note: 500 replications over a five-year period; random effects fixed at 0

performance has relatively little impact on the amount of time predicted to bespent in unemployment and its main effect is to generate a substitution offormal education for employment and YTS training

There is a moderate estimated effect of physical and social disadvantage.Individuals identified as having some sort of (subjectively defined) major healthproblem are predicted to spend a greater proportion of their first five post-schoolyears in college or YTS (43 % rather than 36 %) compared with the otherwisesimilar base case This displaces employment (52 % rather than 57 %), but alsoreduces the time spent unemployed by about two and a half percentage points Inthis sense, there is evidence that the youth employment system was managing toprovide effective support for the physically disadvantaged, if only temporarily.After controlling for other personal characteristics, there is a significant role for

local social influences as captured by the occupational, educational and housingcharacteristics of the local area, and the quality of the individual's school Poorschool and neighbourhood characteristics are associated with a slightly reducedprediction of time spent in college and employment, with a correspondingincrease in unemployment and YTS tenure Nevertheless, compared with raceand education effects, these are minor influences.

16.4.5 The effects of unobserved heterogeneity

To analyse the effects of persistent heterogeneity specific to each state of origin,

we conduct simulations similar to those presented in the previous paragraph.The results are shown in Figures 16.9±16.12 We consider the representativeindividual and then conduct the following sequence of stochastic simulations.For each state k ˆ C, E, U, YTS set all the heterogeneity terms to zero exceptfor one, uk, whose value is varied over a grid of values in the range [ÿ2, 2](covering approximately four standard deviations) At each point in the grid,

500 five-year work histories are simulated stochastically and the average portion of time spent in each state is recorded This is done for each of the four

pro-uk, and the results plotted The plots in Figures 16.9±16.12 show the effect ofvarying each of the heterogeneity terms on the proportion of time spentrespectively in college, employment, unemployment and unemployment

Figure 16.9 The effect of each state-specific random effect on the proportions of timespent in college

Figure 16.10 The effect of each state-specific random effect on the proportions of timespent in employment.

Figure 16.11 The effect of each state-specific random effect on the proportions of timespent in unemployment

The striking feature of these plots is the large impact of these persistentunobservable factors on the average proportions of the five-year simulationperiod spent in each of the four states This is particularly true for college,

Figure 16.12 The effect of each state-specific random effect on the proportions of timespent in YTS.

where the proportion of time spent in education falls from over 20 % at uCˆ 0

to almost zero at uCˆ 2, with a corresponding rise in the time spent inemployment and unemployment The proportion of time spent unemployed(essentially the unemployment rate among individuals of the representativetype) is strongly influenced by all four state-specific random effects, with a 6percentage point variation in the unemployment rate

16.5 SIMULATIONS OF THE EFFECTS OF YTS simulations of the effects of yts

We now bring out the policy implications of the model by estimating theaverage impact of YTS for different types of individual, again using stochasticsimulation as a basis (Robins, Greenland and Hu (1999) use similar tools toestimate the magnitude of a causal effect of a time-varying exposure) A formalpolicy simulation can be conducted by comparing the model's predictions intwo hypothetical worlds in which the YTS system does and does not exist Thelatter (the `counterfactual') requires the estimated model to be modified in such

a way that YTS spells can no longer occur The results and the interpretationalproblems associated with this exercise are presented in Section 16.5.2 below.However, first we consider the effects of YTS participation and and of earlydropout from YTS, by comparing the simulated labour market experience ofYTS participants and non-participants within a YTS world For this we use themodel as estimated, except that the `risk' of attrition (transition to state O) isdeleted

16.5.1 The effects of YTS participation

We work with the same set of reference individuals as in Sections 16.4.4±16.4.5above Again, the state-specific random effects are fixed at their median values

of 0, so that the simulations avoid the problems of endogenous selection arisingfrom persistent unobservable characteristics This time the 500 replications aredivided into two groups: the first one contains histories with no YTS spell andthe second one histories with at least one YTS spell We have then two groups

of fictional individuals, identical except that the first happen by chance to haveavoided entry into YTS, while the second have been through YTS: thus wecompare two potential types of work history, only one of which could beobserved for a single individual (Holland, 1986)

To make the comparison as equal as possible, we take the last three years ofthe simulated five-year history for the non-YTS group and the post-YTS period(which is of random length) for the YTS group We exclude from each groupthose individuals for whom there is a college spell in the reference period, thusfocusing attention solely on labour market participants

Figure 16.13 shows, for the base case individual, the difference in simulatedunemployment incidence for the two groups At the median value of the randomeffects, the difference amounts to approximately 5 percentage points, so thatYTS experience produces a substantially reduced unemployment risk We haveinvestigated the impact of unobservable persistent heterogeneity by repeatingthe simulations for a range of fixed values for each of the uk Figure 16.13 shows

the plot for uU; broadly similar patterns are found for the other uk, suggestingthat the beneficial effect of YTS participation is more or less constant acrossindividuals with differing unobservable characteristics.

Table 16.3 shows the influence of observable characteristics, summarising theresults of simulations for the base case and peturbations with respect to ethni-city, education and area/school quality The beneficial effects of YTS partici-pation are evident in all cases, but are particularly strong for members of ethnicminorities and for those with better levels of school examination achievement.Note that these are the groups with the highest probabilities of full-term YTSspells

16.5.2 Simulating a world without YTS

The ultimate aim of this type of modelling exercise is to say something aboutthe economic effects of implementing a training/employment subsidy schemesuch as YTS The obvious way to attempt this is to compare simulations of themodel in two alternative settings: one (the `actual') corresponding to the YTSscheme as it existed during the observation period; and the other (the `counter-factual') corresponding to an otherwise identical hypothetical world in whichYTS does not exist There are well-known and obvious limits on what can beconcluded from this type of comparison, since we have no direct way ofknowing how the counterfactual should be designed Note that this is not a

Table 16.3 Simulated effects of YTS participation on employment frequency andduration for hypothetical individuals

Replications with no YTS spell Simulated individual % period in work % spells in work

1±3 GCSEs at grade C 85.1 83.0

> 3 GCSEs at grade C 88.3 85.4

Low school and area quality 86.5 84.0

Replications containing a YTS spell Simulated individual % post-YTSperiod in work % post-YTSspells in work Mean YTSduration % YTS spellsfull term

1±3 GCSEs at grade C 96.5 91.8 1.62 63.6

> 3 GCSEs at grade C 98.6 96.8 1.75 73.1

Low school and area quality 90.1 81.1 1.47 51.8

Note: 500 replications over a five-year period; random effects fixed at 0.

problem specific to the simulations presented in this chapter; any attempt togive a policy-oriented interpretation of survey-based results is implicitly subject

to the same uncertainties

The design of a counterfactual case requires assumptions about three majorsources of interpretative error, usually referred to, rather loosely, as deadweightloss, displacement and scale effects Deadweight loss refers to the possibilitythat YTS (whose objective is employment promotion) may direct some re-sources to those who would have found employment even without YTS.Since YTS has some of the characteristics of an employment subsidy, this is astrong possibility It seems likely that if YTS had not existed during ourobservation period, then some of those who were in fact observed to participate

in YTS would have been offered conventional employment instead, possibly onold-style private apprenticeships Displacement refers to a second possibilitythat a net increase in employment for the YTS target group might be achieved

at the expense of a reduction in the employment rate for some other group,presumably older, poorly qualified workers Note, however, that displacementeffects can also work in the other direction For example, Johnson and Layard(1986) showed, in the context of a segmented labour market with persistentunsatisfied demand for skilled labour and unemployment amongst unskilledworkers, that training programmes can simultaneously produce an earningsincrease and reduced unemployment probability for the trainee (which might bedetected by an evaluation study) and also make available a job for one of thecurrent pool of unemployed A third interpretative problem is that the aggre-gate net effect of a training programme may be non-linear in its scale, so thatextrapolation of a micro-level analysis gives a misleading prediction of theeffect of a general expansion of the scheme This mechanism may work, forinstance, through the effect of the system on the relative wages of skilled andunskilled labour (see Blau and Robins, 1987)

The evidence on these effects is patchy Deakin and Pratten (1987) giveresults from a survey of British employers which suggests that roughly a half

of YTS places may have either gone to those who would have been employed

by the training provider anyway or substituted for other types of worker (withdeadweight loss accounting for the greater part of this inefficiency) However,other authors have found much smaller effects (see Jones, 1988), and the issueremains largely unresolved Blau and Robins (1987) found some empiricalevidence of a non-linear scale effect, by estimating a significant interactionbetween programme size and its effects The need for caution in interpretingthe estimated effects of YTS participation is evident, but there exists no clearand simple method for adjusting for deadweight, displacement and scale effects.The economic assumptions we make about the counterfactual have a directparallel with the interpretation of the statistical transition model (see alsoGreenland, Robins and Pearl (1999) for a discussion on this) To say anythingabout the effects of removing the YTS programme from the youth labourmarket requires some assumption about how the statistical structure wouldchange if we were to remove one of the possible states The simulations wepresent in Table 16.4 correspond to the very simplest counterfactual case and,

Table 16.4 Simulated work histories for hypothetical individuals with and without theYTS scheme in existence.

Note: 500 replications over a five-year period; random effects fixed at 0

equivalently, to the simplest competing risks interpretation In the non-YTSworld, we simply force the transition intensities for movements from any stateinto YTS, and the probability of YTS as a first destination, to be zero Theremainder of the estimated model is left unchanged, so that it generates transi-tions between the remaining three states In other words, we interpret the model

as a competing risks structure, in which the YTS `risk' can be removed withoutaltering the levels of `hazard' associated with the other possible destinationstates This is, of course, a strong assumption and avoids the issue of the macro-level effects which might occur if there really were an abolition of the wholestate training programme

As before, we work with a set of hypothetical individuals, and remove theeffect of inter-individual random variation by fixing the persistent individual-specific random effects at zero Table 16.4 then summarises the outcome of 500replications of a stochastic simulation of the model The sequence of pseudo-random numbers used for each replication is generated using a randomlyselected seed specific to that replication; within replications, the same pseudo-random sequence is used for the actual and counterfactual cases Note that theresults are not directly comparable with those presented in Section 16.4.2 whichcompared YTS participants and non-participants, since we are considering here

the whole five-year simulation period rather than the later part of it We arealso not focusing exclusively on the labour market, since we retain in theanalysis individuals who are predicted by the simulations to remain in educa-tion A third major difference is that the analysis of Section 16.4.2 did notconsider the effects of differences in YTS participation frequency together withthe effects of YTS per participant, whereas the simulations reported here willnecessarily show bigger impacts of abolition for groups with high YTS partici-pation rates.

On the basis of these results in Table 16.4, the effect of the YTS programme

on employment frequencies is important but moderate: a fall of no more than 5percentage points in the proportion of time spent in employment Instead, themajor impact of abolition is on time spent in education and in unemployment.With YTS abolished, the proportion of time spent in unemployment rises formost cases by between 6 and 14 percentage points, although the rise is neces-sarily much smaller for those with low probabilities of YTS participation(notably non-whites and those with good GCSE results) The simulated degree

of substitution between continuing education and YTS is substantial, with theduration rising by 4±9 percentage points in every case The rise is largest forindividuals disadvantaged by ethnicity, health or social/educational back-ground, but also for those with a modestly increased level of school examin-ation achievement relative to the base case

16.6 CONCLUDING REMARKS concluding remarks

We have estimated a large and highly complex transition model designed toaddress the formidable problems of understanding the role played by govern-ment training schemes in the labour market experience of school-leavers Thequestion `what is the effect of YTS?' is a remarkably complex one, and we havelooked at its various dimensions using stochastic simulation of the estimatedmodel Abstracting from endogenous (self-)selection into YTS, we have foundevidence suggesting a significant improvement in subsequent employment pro-spects for those who do go through YTS, particularly in the case of YTS

`stayers' This is a rather more encouraging conclusion than that of Dolton,Makepeace and Treble (1994), and is roughly in line with the earlier appliedliterature, based on less sophisticated statistical models Our results suggestthat, for the first five years after reaching school-leaving age, YTS appearsmainly to have absorbed individuals who would otherwise have gone intounemployment or stayed on in the educational system The employment pro-motion effect of YTS among 16±21-year-olds might in contrast be judgedworthwhile but modest Our estimated model is not intended to have any directapplication to a period longer than the five-year simulation period we haveused However, arguably, these results do give us some grounds for claiming theexistence of a positive longer term effect for YTS The increased employmentprobabilities induced by YTS naturally occur in the late post-YTS part of the

five-year history we have simulated As a result, we can conclude that, tional on observables and persistent unobservable characteristics, a greaterproportion of individuals can be expected to reach age 21 in employment, ifYTS has been available during the previous five years, than would otherwise bethe case On the reasonable assumption of a relatively high degree of employ-ment stability after age 21, this suggests a strong, positive, long-term effect ofYTS on employment probabilities.

Table A1 Variables used in the models

Time-invariant characteristics (mean over all individuals):

GCSE2 Dummy for at least 1 General Certificate of Secondary

GCSE3 Dummy for 1±3 GCSE passes at grade C or better 0.185GCSE4 Dummy for at least 4 GCSE passes at grade C or better 0.413ILL Dummy for the existence of a major health problem 0.012SCHOOL Measure of school quality ˆ proportion of pupils with at

least 5 GCSE passes in first published school leaguetable

0.384

AREA Measure of social background ˆ proportion of homes in

ward of residence that are owner-occupied 0.779Spell-specific variables (mean over all episodes):

DATE Date of the start of spell (years since 1 January 1988) 1.11YTSYET Dummy for existence of a spell of YTS prior to the

YTSDUR Total length of time spent on YTS prior to the current

YTSLIM Dummy ˆ 1 if two-year limit on YTS was reached prior

YTSMATCH Dummy ˆ 1 if current spell is in employment and there

was a previous YTS spell in the same industrial sector 0.121CLERICAL Dummy ˆ 1 if current spell is in clerical employment 0.036TECH Dummy ˆ 1 if current spell is in craft/technical

Table A2 Sample transition frequencies (%).

(a) Initial spell

Mean elapsed duration for both

Table A4 Estimates: initial state logit component (standard errors in parentheses)

Table A5(a) Estimates: transition component (standard errors in parentheses).

Coefficient Destination-specific transition intensities

Constant ÿ1.377 (1.85) ÿ5.924 (0.57) ÿ1.820 (0.71) ÿ6.926 (0.78) ÿ4.481 (1.11) DATE ÿ8.090 (2.99) Ð Ð 0.795 (0.11) ÿ1.884 (0.14)

YTSDUR Ð 0.762 (0.18) ÿ1.328 (0.46) ÿ0.198 (0.17) Ð YTSLIMIT Ð ÿ2.568 (0.71) ÿ3.234 (0.75) Ð Ð

E Ð Ð 4.782 (0.59) ÿ0.962 (0.61) 3.469 (1.05)

U 5.530 (0.73) 6.654 (0.54) Ð 5.079 (0.58) 6.066 (1.04) YTS Ð 3.558 (0.49) 2.853 (0.41) Ð Ð U* (GCSE3/4) ÿ0.447 (0.72) 0.927 (0.22) Ð 1.197 (0.36) 1.635 (0.45) SCHOOL Ð 0.233 (0.32) ÿ0.690 (0.47) ÿ1.389 (0.45) 1.451 (0.50) AREA 1.512 (1.23) Ð ÿ1.628 (0.51) Ð Ð URATE Ð ÿ3.231 (1.99) ÿ2.630 (2.62) 8.488 (3.32) 5.724 (3.20) College m1 0.817 (0.16)

s k`

C Ð 2.494 (0.62) 1.171 (0.47) 0.555 (0.26) 5.547 (1.48)

E 0.414 (1.70) Ð 4.083 (0.43) 0.555 (0.26) 5.465 (0.75)

U 2.429 (0.41) 1.652 (0.13) Ð 0.555 (0.26) 1.508 (0.12) YTS 5.569 (4.45) 1.018 (0.36) 1.315 (0.40) 0.555 (0.26) Ð

Table A6 Estimates: YTS limit logit (standard errors in parentheses).

PART E

Incomplete Data

Analysis of Survey Data Edited by R L Chambers and C J Skinner

Copyright ¶ 2003 John Wiley & Sons, Ltd.

ISBN: 0-471-89987-9

``missingness'' Skinner and Holmes provide an illustration of this in Chapter

14 where variables in different waves of a longitudinal survey are recorded fordifferent numbers of units because of attrition

The three chapters making up Part E of this book explore survey data analysiswhere relevant data are, in one way or another, and to different extents, missing.The first, by Little (Chapter 18), develops the Bayesian approach to inference ingeneral settings where the missingness arises as a consequence of different forms

of survey nonresponse In contrast, Fuller (Chapter 19) focuses on estimation of

a finite population total in the specific context of two-phase sampling This is anexample of ``missingness by design'' in the sense that an initial sample (the first-phase sample) is selected and a variable X observed A subsample of these units isthen selected (the second-phase sample) and another variable Y observed for thesubsampled units Values of Y are then missing for units in the first-phase samplewho are not in the second-phase sample As we show in the following section, thischapter is also relevant to the case of item nonresponse on a variable Y whereimputation may make use of information on another survey variable X Finally,

¶

ISBN: 0-471-89987-9

Steel, Tranmer and Holt (Chapter 20) consider the situation where either the unitlevel data of interest are missing completely, but detailed aggregates correspond-ing to these data are available, or a limited amount of survey information isavailable, but not sufficient to allow linkage of analysis and design variables inthe modelling process.

In Chapter 18 Little builds on the Bayesian theory introduced in Chapter 4 todevelop survey inference methods when data are missing because of nonre-sponse Following standard practice he differentiates between unit nonre-sponse, where no data are available for survey non-respondents, and itemnonresponse, where partial data are available for survey non-respondents Inthe first case (unit nonresponse), the standard methodology is to compensatefor this nonresponse by appropriately reweighting the responding sample units,while in the second case (item nonresponse) it is more usual to pursue animputation strategy, and to fill in the ``holes'' in the rectangular completedata matrix by best guesses about the missing variable values From a theoret-ical perspective, this distinction is unnecessary Unit nonresponse is just anextreme form of item nonresponse, and both weighting (or more generally,estimation) and imputation approaches can be developed for this case InSection 17.3 below we briefly sketch the arguments for either approach andrelate them to the Bayesian approach advocated by Little

Fuller's development in Chapter 19 adopts a model-assisted approach todevelop estimation under two-phase sampling, using a linear regression modelfor Y in terms of X to improve estimation of the population mean of Y Sincetwo-phase sampling is a special case of item nonresponse, the same methodsused to deal with item nonresponse can be applied here In Section 17.4 below

we describe the approach to estimation in this case Since imputation-basedmethods are a standard way of dealing with item nonresponse, a naturalextension is where the second-phase sample data provide information forimputing the ``missing'' Y-values in the first-phase sample

The development by Steel, Tranmer and Holt (Chapter 20) is very different inits orientation Here the focus is on estimation of regression parameters, andmissingness arises through aggregation, in the sense that the survey data, to agreater or lesser extent, consist not of unit-level measurements but of aggregatemeasurements for identified groups in the target population If these group-level measurements are treated as equivalent to individual measurements thenanalysis is subject to the well-known ecological fallacy (Cleave, Brown andPayne 1995; King, 1997; Chambers and Steel, 2001) Steel, Tranmer and Holt infact discuss several survey data scenarios, corresponding to different amounts

of aggregated data and different sources for the survey data, including ations where disaggregated data are available from conventional surveys andaggregated data are available from censuses or registers Integration of surveydata from different sources and with different measurement characteristics (inthis case aggregated and individual) is fast becoming a realistic practical optionfor survey data analysts, and the theory outlined in Chapter 20 is a guide tosome of the statistical issues that need to be faced when such an analysis isattempted Section 17.5 provides an introduction to these issues