Skip Header

Working Paper Number SEHSD-WP2015-19
Ashley Edwards


Over the period from 2009 to 2012, the median length of a poverty spell was 6.2 months. While longitudinal data over multiple years has the benefit of capturing a greater number of poverty spells and reducing the proportion of spells affected by left or right-censoring, researchers and policymakers must wait years for data to be released, and the impact of sharp swings in the economy is masked over extended periods.

The biggest limitation in providing single-year estimates of poverty spell dynamics relates to issues of spell capture and censoring. As researchers shorten the observation period, they capture fewer poverty spells and a greater proportion of spells are unable to be assigned “true” start and end dates as spells begin prior to the start of the observation period or continue beyond the end of the observation period.

Analysis within single calendar years from 2009 to 2012 captures approximately 20 to 35 percent fewer spells when compared to the 4-year period, and left-censoring increases by approximately 240 to 265 percent. Further, the traditional practice of dropping left-censored poverty spells from duration estimates introduces an unexplored amount of bias, as approximately 55 to 60 percent of observed spells are excluded due to left-censoring.

This research presents estimates of single-year poverty dynamics across various demographic groups over the period 2009 to 2012. Findings suggest that small differences in medians across calendar years combined with relatively high standard errors limit the conclusions that can be drawn by comparing year-to-year changes. Additionally, while the characteristics of left-censored spells captured within a single calendar year are comparable with left-censored spells observed over the 4-year period, the increased capture of left-censored spells within single calendar years leads to greater bias when excluding left-censored spells, and often prohibits the assignment of median spell durations.

Back to Header