Nonresponse in BLS Housing Survey biases CPI shelter inflation
A new working paper examines how nonresponse in the BLS Housing Survey affects the accuracy of US CPI shelter inflation. Researchers find that current imputation methods, while potentially biased, yield similar inflation indexes in practice.
Nonresponse challenges CPI shelter data
Shelter is the largest component of US consumer price index (CPI) inflation, making its accurate measurement critical for overall inflation figures.
The BLS Housing Survey, which underpins CPI shelter inflation, has seen nonresponse rates increase significantly since 2000, now affecting roughly 40 percent of total observations.
This growing issue is concerning because missing rent data are currently imputed using a class-mean approach based on rent tier, which the authors find may result in biased imputations.
Their analysis reveals that nonresponse is correlated with factors beyond just rent tier, such as structure type and tenure length, and that inflation rates vary between imputed and non-imputed units.
Even minor inaccuracies in shelter indices can lead to economically meaningful distortions in overall inflation, underscoring the urgency of addressing these measurement issues.
Alternative imputations, similar outcomes
The paper explores alternative simple imputation methods that incorporate variables like structure type and tenure length, which are strongly correlated with both nonresponse and rent growth.
While a simple model suggests these alternatives could theoretically yield different index biases, the study's practical application shows they produce similar shelter inflation indexes.
This indicates that any actual index bias from current methods may be modest.
The authors contribute by highlighting the increasing problem of survey nonresponse in official statistics, analyzing its patterns in the BLS Housing Survey, and detailing the current BLS imputation method.
A critical but contained bias
The study confirms a long-suspected issue with CPI shelter data, providing valuable empirical insight into imputation biases.
While the theoretical risk of bias is high, the practical impact appears modest, suggesting current methods are more robust than feared.
This offers reassurance for policymakers relying on CPI data, though continued vigilance on survey quality remains essential.