ASECWT in 2014 ASEC

CPS
1

I am wondering whether there is an error with the ASECWT variable in the 2014 ASEC data. Using the weight to create time-series statistics leads to some odd results. As a check, I calculated means for asecwt for several years and the 2014 weight is much larger than adjacent years (the following is Stata output):
. mean asecwt, over(year)

Mean estimation Number of obs = 1,948,983

          |       Mean   Std. Err.     [95% Conf. Interval]

--------------±-----------------------------------------------
c.asecwt@year |
2010 | 1450.328 1.964177 1446.478 1454.178
2011 | 1495.506 2.066882 1491.455 1499.557
2012 | 1533.418 2.152923 1529.198 1537.637
2013 | 1535.36 2.116205 1531.212 1539.508
2014 | 3141.166 6.070837 3129.268 3153.065
2015 | 1588.592 2.185418 1584.309 1592.875
2016 | 1719.088 2.474937 1714.237 1723.939
2017 | 1723.227 2.479121 1718.368 1728.086
2018 | 1794.474 2.614394 1789.35 1799.598
2019 | 1800.966 2.595118 1795.88 1806.053

2

I recommend reading this blog post and this page for more information on this sample. In short, there are two sub-samples in the 2014 sample, each of which are weighted to be nationally representative. One method not included in that post for adjusting the weights with pooled data is to scale the weights based on the size of the subsample:

HFLAG = 0 indicates the 5/8 file – their weights should be multiplied by 5/8

HFLAG = 1 indicates the 3/8 file – their weights should be multiplied by 3/8

1 Like