What the mean wage hides

The mean wage is the most quoted figure about any occupation, and the least informative. Mean versus median covers the arithmetic. This page is about everything else the mean leaves out, and most of it applies to the median too.

Skew

An average is pulled toward whichever tail is longer. Wages have a floor and no ceiling, so the long tail is at the top, and the mean exceeds the median for nearly every occupation. For an occupation with a few very highly paid members the gap can be large enough that a majority of workers earn less than “the average.” The percentile table, not the mean, shows where the bulk of workers sit; see reading wage percentiles.

Who is not in it

OEWS counts wage and salary workers in nonfarm establishments. It excludes the self-employed, partners and owners of unincorporated businesses, and private household workers. For occupations where self-employment is a large share, such as real estate brokers, photographers, writers, hairdressers and many construction trades, the mean describes only the employed part of the occupation and can differ substantially from what the typical practitioner earns.

What pay is counted

The wage is straight-time, gross pay. It includes base pay, commissions, tips and production bonuses, and excludes overtime premiums, shift differentials, holiday and year-end bonuses, stock, severance and benefits. An occupation whose income is heavily overtime, such as some police and nursing jobs, or heavily equity, such as some technology roles, has a mean that understates total compensation. An occupation with generous benefits and modest cash pay, such as many public-sector jobs, looks lower against private-sector peers than a total compensation comparison would show.

Hours

The hourly mean is a rate, and the annual mean is that rate times 2,080 hours. Neither says how many hours workers actually get. An occupation with many part-time workers can have a respectable hourly mean and modest actual annual earnings; see annual versus hourly figures.

Mixed content inside one code

A SOC code groups jobs by the work performed, but the grouping can be broad. “Managers, all other” or “software developers” contains people doing quite different work at quite different levels. The mean of such a code averages across the mix, and a change in the mix between releases moves the mean with no change in anyone’s pay. Residual “all other” codes are the clearest case.

Place and industry

A national mean averages across every state and industry. Within it, the same occupation can have a mean in one metro a third higher than in another; see why metro wages differ. The national figure is right for a national question and wrong for a local one. The state and metro tables exist so that the local figure is available where BLS publishes it.

Measurement

Every mean is an estimate from a sample, with a published relative standard error, and the mean for occupations with top-coded tails depends on an external estimate of the top interval; see top-coded wages. Small differences between two means, or between two years, are often within the error.

What to use instead

Start with the median. Read the 25th and 75th percentiles for the middle half. Check the 10th against the minimum wage in the state to see how close the floor is. Note employment, because an estimate from a few hundred workers is noisier than one from a few hundred thousand. Then read the mean, as a measure of how far the top pulls the average, and not as the answer to “what does this job pay.”

The mean is not wrong. It is one number summarizing a distribution, chosen for arithmetic convenience, and the occupation pages on this site show it only beside the five numbers that describe the distribution itself.