The redesigned 2014 SIPP instrument allows respondents to report wage and salary income in several different ways. In particular, respondents are allowed to report an hourly pay rate, a monthly pay rate, an annual salary, or several other alternatives. Respondents are also able to report up to two changes in their pay rate over the year (e.g., pay raises or pay cuts).

While we believe that this flexibility in reporting generally increases data quality by allowing respondents to report in the way that is most salient and accurate for their situation, it can also lead to some mistakes. In particular, someone may report (perhaps accidentally) that they were paid an hourly wage, and then report a monthly pay rate or annual salary instead of reporting that hourly wage. As a hypothetical example, someone could report that he was paid an hourly wage, and when asked the amount of the hourly wage he could say $40,000. This is likely the income that he earned over the entire year as opposed to the income that he earned per hour worked. In order to mitigate this problem, a window will appear with the following text: “This amount seems unusually high. Are you sure this is the correct amount of your hourly rate?” Similarly, a respondent could report that he was paid an annual salary, and report that the amount was $10. This is likely the income that he earned per hour worked rather than the income that he earned over the entire year. Checks similar to the one noted above come up for all amounts flagged as particularly high or particularly low for all reported rates of pay. In testing, these “soft checks” helped to correct reporting in most instances. However, there are a few cases when someone verified the particularly high or low amount and moved on with the interview.

Particularly high reported hourly pay rates create more visible issues than other types of misreporting; in processing we multiply these wages by hours worked to construct weekly and monthly earnings. In the hypothetical example above, if the person worked 40 hours per week for 52 weeks of the year, a $40,000 hourly pay rate would imply monthly earnings of about $7 million and annual earnings of about $83 million. When particularly high or particularly low pay rates were reported and subsequently verified we elected to leave these amounts as reported. Although we can infer that the type of pay was likely misreported, it is many times unclear what type of pay the respondent should have reported. It is worth noting, however, that particularly high and low amounts are excluded from imputation procedures. On the public use data, annual salaries that were reported as hourly pay rates will be top-coded and replaced with the median value of the distribution above the top-code. The mean of the distribution above the top-code is also available to public data users. However, public data users will not be able to distinguish between censored values that are likely hourly pay rates and censored values that are likely a different type of pay such as an annual salary.

Users who would like to do analysis related to the mean of wages or earnings (such as regression analysis or aggregate earnings statistics), have several options to deal with outliers stemming from misreported pay frequency. However, analysis should be done with caution and potentially on a case-by-case basis and in concert with other reported earnings such as commission, tips, overtime, bonus payments, and business profits. Also, while high values are more salient, particularly low values should be considered as well.

Exclude the outliers.
Delete the reported hourly wage and impute a new value for topcoded wages.
Recode the value of EJB(n)_PAYHR, and associated earnings recodes as necessary. This will be difficult for the values that are top-coded on the public use file. However, it should be easier for those who reported what looks like an hourly pay rate as an annual salary, for example.

Variables that are affected by high hourly wages being multiplied: