Current unemployment levels could be underestimated by up the 0.5%
Because the source of the unemployment rate is a sample of around 15,000 households which are compulsorily surveyed.
Naturally enough some people refuse to disclose information, can't be contacted, move, etc.To adjust for this the surveyors weight households with similar characteristics to those not responding. To test how effective this is researchers matched March 2013 HLFS with Census 2013 data.
The results suggest that the calibration employed by the HLFS does a reasonable job of adjusting for non-response (once region is included as a benchmark), and that further non-response adjustments have minimal effect on reducing the bias introduced by non-response.However,
Using the linked dataset we discovered that bias from non-response exists in the estimates of labour force status from the HLFS. The unemployment rate in the responding sample is underestimated by around 0.3 percent to 0.5 percent, while the labour force participation rate is underestimated by around 0.2 percent. Unemployed people had both the highest non-contact rate and the highest refusal rate. This differs from other studies’ findings, where employed people were found to be harder to contact (Foster, 1998; Stoop, 2005).
So unemployment could be as high as 6.2%
This problem has always existed but appears to be worsening.
Response rates for the HLFS were around 86 percent on average over the four years to 2014, but there has been a downward trend in response rates in recent years that reached an all-time low in the June 2013 quarter (80.8 percent).
Still, many countries use a form of the HLFS to assess labour force status and will no doubt have similar problems. Arguably comparability and tracking trends is more important than point-in-time accuracy.