What Efficacy Is, and Why It Stopped Being the Question

Why this matters

Efficacy is the number every product sheet leads with and the number every rebate conversation starts from, and for about a decade it was the right thing to lead with, because the spread between technologies was a factor of six. It is not that anymore. The spread between two current candidates is closer to a factor of 1.2, while the spread in how much of their output reaches the task can be a factor of 1.4 in the same room. Pick on the headline and you can install the more efficient luminaire and burn more watts per delivered footcandle, which is the outcome nobody checks for because the specification never asked the question in those terms.

Both terms of the ratio carry conditions

Efficacy = lumens out / watts in, and each half is only meaningful with its condition attached.

The numerator is the finished luminaire's delivered output from an absolute photometric and electrical test, not the LED package's rating and not a family maximum. The gap between those is not small; the cut-sheet card walks a case where it is 29 percent.

The denominator is the luminaire's input watts at the tested drive current, measured at the same time as the lumens, including driver losses. It is not the LED module's wattage, and it is emphatically not an "equivalent wattage" claim, which is a marketing comparison to a product nobody is buying. Where a field verification of connected load is genuinely needed, that is energized electrical work for a qualified person under 29 CFR 1910.333(b), inside the shock and arc-flash boundaries of NFPA 70E-2021 in the edition your employer's electrical safety program adopts; for a specification comparison, the test report already has the number and nobody needs to open a panel.

Both terms also move with configuration. The same family at a higher color temperature and lower color rendering will publish a higher efficacy than the code you are actually buying, because those two settings raise lumens per watt at the cost of things the color cards cover.

The number stopped discriminating because the spread collapsed

Rough magnitudes, worth carrying as ranges rather than as figures, and routed to the manufacturer's test report for anything you are quoting:

When the choice was incandescent against fluorescent, efficacy alone settled it and nothing else needed checking. Between two LED luminaires from the current market, efficacy is usually a difference in the teens of percent, which is smaller than the difference the room can make. So the question moved. It is no longer "how efficient is the source" but "how much maintained light lands on the plane per watt installed", and that quantity has a name worth using:

Delivered efficacy = source efficacy x coefficient of utilization x light loss factor, in the same lumens-per-watt unit, for a stated room, a stated plane and a stated maintenance basis.

Inside one family, efficacy moves for reasons that are not quality

Before comparing two manufacturers, understand that the same LED module produces different efficacy figures depending on how the luminaire is built around it, and the directions are predictable:

• Higher color temperature publishes higher efficacy than lower, at the same drive, because of how the phosphor conversion works. That is a real gain and it is paid for on the color side.
• Lower color rendering publishes higher efficacy than higher, for the same reason. Also real, also paid for elsewhere.
• Harder drive current lowers efficacy while raising output. LED output rises sublinearly with current and the junction runs hotter, so lumens per watt falls as the module is pushed.

That last one has a consequence worth stating plainly, because it inverts the usual reading. Take one module at half its rated current: input power drops to roughly 50 percent while output holds at roughly 55 percent of rated, so efficacy rises by a factor of 1.10. Nothing about the module improved. It is being asked to do less.

So two luminaires built on the same module, one driven softly and one driven hard, will publish different efficacies, and the softly-driven one wins on the sheet while producing fewer lumens per luminaire and therefore needing more luminaires to hit the same target. A higher efficacy figure can be a de-rated product, and the only way to tell is to read output and input power as a pair rather than reading their quotient. This is the same direction as the 70 percent drive current yielding 75 percent of flux that the photometric file card walks, so the two figures are consistent parts of one curve.

The gate

Does a difference in source efficacy predict a difference in energy per unit of delivered light?

Only when the coefficient of utilization and the light loss factor are the same for both candidates. Where they are, the two extra factors are a common multiplier and the source-efficacy ratio passes through untouched. Where they are not, the ratio is meaningless and can invert. Two cases, same two products.

Candidate P: 13,200 lm at 100 W, so 132.0 lm/W source efficacy. Aisle-optimized distribution. Candidate Q: 13,320 lm at 90 W, so 148.0 lm/W source efficacy. General wide distribution.

Q leads on the headline by 148.0 / 132.0 = a factor of 1.121, or 12.1 percent.

Outcome one: the storage aisle, gate closed

A tall narrow storage bay with a high room cavity ratio and dark racking. Coefficients of utilization read from each manufacturer's table for this room geometry and reflectance triple: P = 0.64, Q = 0.46, because P's distribution throws down the aisle and Q's throws into the racks. Light loss factor 0.80 for both, since the environment and cleaning interval are identical.

• P delivered efficacy = 132.0 x 0.64 x 0.80 = 67.6 lm/W
• Q delivered efficacy = 148.0 x 0.46 x 0.80 = 54.5 lm/W

P now leads by 67.6 / 54.5 = a factor of 1.24, or 24 percent, on a target of maintained horizontal illuminance at the floor of the aisle. The rank flipped, and it flipped by more than the headline gap it reversed. Specify Q here and you install the luminaire with 12 percent better lumens per watt and spend 24 percent more watts for the same maintained light.

Outcome two: the open office, gate open

Same two products, low room cavity ratio, light surfaces, and both distributions appropriate to the space, so the CU tables give 0.68 for both and the cleaner environment gives 0.85 for both.

• P delivered efficacy = 132.0 x 0.68 x 0.85 = 76.3 lm/W
• Q delivered efficacy = 148.0 x 0.68 x 0.85 = 85.5 lm/W

Q leads by 85.5 / 76.3 = a factor of 1.121, which is the source-efficacy ratio to three digits, exactly as the gate predicts, because 0.68 x 0.85 is a common factor that cancels out of the comparison. Here the headline was a valid proxy, and it was valid for a reason you can state rather than by luck.

The term neither case contains

Both outcomes compare connected load for equal delivered light. Neither says anything about energy consumed, because that is connected load multiplied by hours at each output level, and controls move that term further than either product difference above. A space with occupancy sensing and daylight response can spend a large fraction of its scheduled hours below full output, and the two candidates will not respond identically: one may dim smoothly to a low floor and the other bottom out well above it, which changes how much of the daylight harvest is actually available.

That is a measured quantity, not a catalog one. It comes from the dimming performance data for the driver and from metered operating hours in the space, and it is the one place where asking the manufacturer a direct question beats any comparison arithmetic. What flips the recommendation: on a 24-hour operation with no daylight and no vacancy, the control term goes to 1.0 for both and the delivered-efficacy comparison above is the whole answer. On a daylit space occupied one shift a day, it is not.

Checking your own figures

• Gate stated, then tested in both directions. Aisle case: CU 0.64 against 0.46, not equal, gate closed, so the headline ratio was refused. Office case: CU 0.68 and 0.68, LLF 0.85 and 0.85, equal, gate open.
• The open-gate case reproduces the headline ratio exactly. 85.5 / 76.3 = 1.121 against the source ratio 148.0 / 132.0 = 1.121. If those two had not matched, the arithmetic would be wrong, because the common factor must cancel.
• The closed-gate case inverts by more than the headline. Headline gap 12.1 percent to Q, delivered gap 24 percent to P, so the flip is not marginal and does not depend on rounding.
• The drive-current effect points the same way in both places it appears. Half current giving 55 percent of flux is an efficacy factor of 1.10; 70 percent current giving 75 percent of flux is a factor of 1.071. Both above 1.00, both rising as drive falls, so the two are on one curve rather than contradicting each other.
• Output and input read as a pair, never as a quotient alone. P is 13,200 lm at 100 W and Q is 13,320 lm at 90 W, so Q's efficacy lead is a wattage difference of 10 W rather than a meaningful output difference of 120 lm.
• Both terms of every efficacy figure sourced. Numerators are tested luminaire output, denominators are tested luminaire input watts at the same drive current. No package rating and no equivalent-wattage figure entered the arithmetic.
• Delivered efficacy carries its conventions. Aisle result is on maintained horizontal illuminance at the aisle floor; office result is on the same basis for that room. They are not compared to each other, because they sit on different planes in different rooms.
• No factor appeared in an example that the sections above did not define. CU, LLF and the two source efficacies are the only inputs, and the control term was named as absent from both cases rather than quietly folded into one.
• Ranges labeled as ranges. The technology table gives orders of magnitude with a route to the test report, and none of its values were used in any calculation.

References

• IES LM-79, the electrical and photometric measurement method for solid-state lighting products, which is where both terms of a defensible efficacy figure come from
• Manufacturer coefficient of utilization tables and driver dimming performance data for the specific ordering codes being compared
• 29 CFR 1910.333(b) for energized electrical work, and NFPA 70E-2021 for the shock and arc-flash boundaries, in the edition your employer's electrical safety program adopts
• See related: Why Lumens and Lux Are Not Interchangeable; How to Read a Luminaire Cut Sheet; What a Photometric File Actually Describes