The use of Cohort Size Shrinkage Index (CSSI) to quantify regional famine intensity during the Chinese famine of 1959-61
There has been a growing interest in studying the causes and impact of the Great Chinese Famine of 1959-61. The Cohort Size Shrinkage Index (CSSI) is the most widely used measure to examine famine intensity and was used in at least 28 Chinese famine studies to date. We examined the potential impact of violations of three requirements for a valid CSSI measure: reliable information on cohort size by year of birth; a stable trend of cohort size by year of birth; and the absence of significant regional migration. We used data from the 1% China 2000 Census to examine the trend of cohort size over time and concentrated on the time window between 1950-70 to exclude policies and events with a large impact on birth trends other than the famine itself. Across China we established a significant difference in cohort size trends between pre-famine births and post-famine births, violating one of the main requirements for a valid CSSI measure. This leads to systematic differences in CSSI depending on what non-famine years are selected for comparison. At the province level, CSSIs estimated based on pre- & post-famine births tend to overestimate famine intensity at higher exposure levels and underestimate intensity at lower levels compared to CSSIs based on pre-famine births alone. This is problematic and demonstrates that the CSSI is not as robust an estimator of famine intensity as had been assumed previously. We recommend therefore that all CSSI should be based on pre-famine birth trends. Using data from Sichuan province, we demonstrate a less pronounced dose-response relation between famine intensity and tuberculosis outcomes using pre-famine based CSSI as compared to reported patterns based on pre- & post-famine based CSSI. We encourage researchers to re-examine their results of Chinese famine studies as local differences in cohort size of pre-famine and post-famine births may lead to significant discrepancies of CSSI estimation and change the interpretation of their findings.