Job Search Friction Index Methodology

The question

How hard is it to find a job right now? Official statistics answer adjacent questions — how many people are unemployed, how many openings exist, how fast employers hire — but none of them answers that one, and a job seeker cannot combine them by eye.

The index is that combination, expressed on a scale where average 2019 conditions read 100. It is an interpretation layer over public data, not new data.

Components

Three dimensions, each oriented so a higher number means a harder search.

Opportunity pressure

How many people are competing for each opening? CPS unemployment level divided by JOLTS job openings, both seasonally adjusted and both in thousands.

Hiring velocity

How quickly are employers actually hiring? JOLTS hires rate, seasonally adjusted, as a percent of employment.

Search duration

How long are people spending unemployed, and how many get stuck? CPS median duration of unemployment in weeks, blended evenly with the share of unemployed people out of work 27 weeks or longer. Both seasonally adjusted.

Search duration blends two series in one slot rather than giving each its own. Median duration and the long-term share correlate at r = 0.93 across the whole series — close enough to the same measurement that separate slots would hand half the index to a single phenomenon.

What was tested and rejected

The quits rate was considered as a fourth dimension and dropped: inverted, it correlates with unemployed-per-opening at r = 0.85, so it mostly restates a dimension already present.

Job-seeker sentiment was deliberately excluded. Confidence surveys measure how people feel about their prospects, which is a different thing from how the market is behaving, and folding them together would make it impossible to say whether a move was conditions or mood.

Averaging the components as simple ratios to their 2019 values was tested first and rejected. Unemployed-per-opening swings across a thirteen-fold range while the other measures move within about two, so that composite tracked it at r = 0.98 and added nothing. The kill test for this project was whether a composite beats unemployed-per-opening alone; the ratio version failed it, and the standardised version passes at r = 0.84.

Calculation

Each dimension is standardised against its own history using a median and a median absolute deviation. A median absolute deviation rather than a standard deviation because April 2020 would otherwise set the scale for the entire series and compress every ordinary month into a narrow band.

The three standardised dimensions are averaged with equal weights. Weighting was tested: doubling any single dimension, or dropping one, produces a series correlating with the equal-weight version at between 0.91 and 0.99. The index is not sensitive to reasonable reweighting, and inventing precise weights would be inventing precision.

The result is rescaled so average 2019 conditions read 100 and one standard deviation of the composite is 15 points. The index is not capped: the Great Recession peak is above 148.

Where this index is wrong

Between 2020-03 and 2021-02 the index is structurally unreliable and should not be used. During a mass layoff followed by mass rehiring, two of the three dimensions invert: median duration collapses because the unemployment pool has just turned over, and the hires rate spikes because furloughed workers are being recalled.

In May 2020 the index reads 72.8 — lower, meaning easier, than the tightest labour market on record. Only the opportunity dimension registered the shock and it was outvoted by the other two.

Those months keep an index value and appear in the chart and the data, marked as unreliable, but they are excluded from the reference distribution used for percentiles. A percentile is a claim that months are comparable, and they are not.

Percentiles and regimes

Percentiles compare a reading against 294 months of history excluding that window. Regime labels are percentile cut-points, not round numbers chosen for how they look:

• Below the 20th percentile — Much easier than typical
• 20th to 40th — Easier than typical
• 40th to 60th — Typical
• 60th to 80th — Harder than typical
• 80th to 95th — Very difficult
• Above the 95th — Extreme friction

An earlier version used 10/30/70/90 cut-points. That put 40% of all history into one band called “around normal”, which described June 2026 and June 2019 identically despite the two sitting 36 percentile points apart. These bands are evenly spaced through the middle and tighten at the top, where the distinctions matter; occupancy across the series runs 20/20/20/20/15/5.

On current data those cut-points fall at index values of 99.9 (20th), 104.8 (40th), 109.5 (60th), 126.1 (80th), 145.1 (95th).

Seasonal adjustment

Every source series is seasonally adjusted. Mixing adjusted and unadjusted series would put a seasonal cycle into some dimensions and not others, and the composite would then partly measure the calendar.

Publication lag

JOLTS publishes roughly a month behind the household survey, so the index always stops one month short of the newest labour-market data. The composite is calculated for the latest month where every source series exists — currently 2026-06. Household data published for later months is shown separately as later signals and is never folded into the index, because that would mix two reference months in a single number.

Revisions

BLS revises its data. Every observation is stored with the date it was retrieved, and the whole series is recalculated when the snapshot is refreshed, so a revised history produces a revised index rather than a frozen first print. The retrieval date is published alongside the index — currently 2026-08-22.

Methodology versioning

This is version 1.0.0, effective 2026-08-22. If the formula changes, the version changes with it and the reason and effect are recorded here. A recurring indicator that silently redefines itself is not an indicator.

Limitations

• National only. State-level JOLTS is published annually rather than monthly, so a monthly state index is not possible from these sources.
• It describes conditions, it does not forecast them. Duration measures describe the existing stock of unemployed people and lag turning points by months.
• It is an average across a labour market that is never uniform. Occupation, industry, seniority and location all move separately, and no national index speaks to an individual search.
• It measures friction for people who are unemployed and looking. It says nothing directly about people searching while employed.

Sources

• BLS JOLTS — Job openings level (series JTS000000000000000JOL)
• BLS JOLTS — Hires rate (series JTS000000000000000HIR)
• BLS Current Population Survey — Unemployment level (series LNS13000000)
• BLS Current Population Survey — Median duration of unemployment (series LNS13008276)
• BLS Current Population Survey — Unemployed 27 weeks and over, share of unemployed (series LNS13025703)

Retrieved from the U.S. Bureau of Labor Statistics public API. The index is our calculation; the underlying statistics are theirs and should be cited as such where the underlying figure is the point.