This article studies the potential measurement errors when coding occupational data. The quality of occupational data is important but often neglected. We recoded open-ended questions on occupation for last and current job in the Dutch SHARE data, using the CASCOT ex-post coding software. The disagreement rate, defined as the percentage of observations coded differently in SHARE and CASCOT, is high even when compared at ISCO 1-digit level (33.7% for last job and 40% for current job). This finding is striking, considering our conservative approach to exclude vague and incomplete answers. The level of miscoding should thus be considered as a lower bound of the “true” miscoding. This highlights the complexity of occupational coding and suggests that measurement error due to miscoding should be taken into account when making statistical analysis or writing econometric models. We tested whether the measurement error is random or correlated to individual or job-related characteristics, and we found that the measurement error is indeed more evident in ISCO-88 groups 1 and 3 and is more pronounced for higher educated individuals and males. These groups may be sorted in occupations that are intrinsically more difficult to be classified, or education and gender may affect the way people describe their jobs.
|Titolo:||Measurement error in occupational coding: an analysis on SHARE data|
|Data di pubblicazione:||2014|
|Appare nelle tipologie:||7.01 Working paper|
File in questo prodotto:
|WP_DSE_belloni_brugiavini_meschi_tijdens_24_14.pdf||N/A||Accesso libero (no vincoli)||Open Access Visualizza/Apri|