将在线招聘广告与概率样本数据相结合增强对职位空缺的小区域估计.pdf

1、WIN Conference.From Web to Data.3-4 February 2025,Gdask,PolandCombining online job advertisementswith probability sample data for en-hanced small area estimation of jobvacanciesDonatas levinskas,Andrius iginas,Ieva BurakauskaitOutlineIntroductionJob vacancy estimation in domainsIntegration of NP sam

2、plesMethodologyThe case of NP based on OJAFurther small area estimation modelingEBLUP based on the FH modelEmpirical resultsEffectiveness for a single quarterEffectiveness for multiple quarters3-4 February 2025,Gdask,PolandWeb Intelligence Network Conference.From Web to Datap.2 of 18Auxiliary inform

3、ation in sample surveysWhat is the main product of National Statistical Institutes(NSIs)?Official statistics.NSIs aim for improvement:by timeliness more frequent estimates,by granularity more detailed level estimates.Typically sample designs are optimized for population-levelestimates.Small domains

4、often have:limited or unplanned sample coveragesmall sample sizeshigh variability or unreliable estimatesA possible solution:incorporate administrative data or othernon-traditional data sources(mobile network,social media,etc.)tosupplement existing probability sample data.3-4 February 2025,Gdask,Pol

5、andWeb Intelligence Network Conference.From Web to Datap.3 of 18Job vacancy data and the problemData sourceTarget variable,yAuxiliary data,xNP sampleP sampleProbability sample data on job vacancies in companies are collectedin the quarterly Statistical survey on earnings.There is complete administra

6、tive information on the monthlynumber of employees,economic activity,etc.Transformed online job advertisement(OJA)data:only partially covers the survey population;as non-probability(or big data)sample is not representative;roughly approximates job vacancies by nonlinear relationship.3-4 February 202