CPI99 Income Conversions

Hello! I am working with ACS data and assigning poverty percentiles based on family income in 2019. My colleague started this work, but is currently out on leave and found an error in how the 2019 FTOINC is calculated. She received this previous instruction from IPUMS staff: “If your analysis requires a consistent family income and poverty measure, you can recalculate FTOTINC by summing INCTOT within each family unit (identified by SERIAL and FAMUNIT), and then recalculating POVERTY by first converting your new family income value to 1999 dollars by multiplying it by CPI99, and then comparing it to the 1999 poverty thresholds.”

When I use the CPI99 value of .652 to convert 2019 dollars to 1999 dollars, the poverty thresholds that are computed do not perfectly match those in the 1999 poverty threshold table: https://usa.ipums.org/usa/volii/poverty.shtml (For example, a family of 5 with 3 children in 2019 has a poverty threshold of $30,510. Converted using CPI99, that threshold in 1999 should be $19,892, but the listed 1999 threshold is $19,882.)
It seems that a small number of households who are under the poverty threshold in 2019 may appear over the 1999 poverty threshold.

Do you know why this conversion error is occurring? Is there another appropriate way to calculate this?

If I understand correctly, the issue you are describing is that multiplying the poverty thresholds provided by the Census Bureau by the CPI-U provided by IPUMS in the table here yields poverty thresholds that are slightly different from those provided by the Census Bureau. The cause of this discrepancy is likely that the CPI-U values provided in the table are rounded to 3 decimal places, while the Census Bureau calculates poverty thresholds using the CPI-U with more decimal places. For better accuracy, I recommend using the poverty thresholds provided by Census for each year rather than converting them using these CPI-U values. Alternatively, by dividing the Census Bureau’s published 1999 poverty threshold by their poverty threshold in another year, you will obtain the exact CPI-U with more decimal places. You can use the resulting CPI-U in your own calculations to ensure the poverty threshold matches what you are seeing in the Census Bureau’s published tables.

Please let me know if I have misunderstood your question.