The numbers below are the ones Masumi Izawa, Michael French and Alan Hedge published in Human Factors in October 2011 from the draft reports and appendices that Charles Snow kept and that are now Collection 5167 at Cornell. They are group averages: Snow recorded output five times a day for each group, but no sheet of individual output survives, and for the third test only the average for each period survives. Where Roethlisberger and Dickson’s 1939 summary, which was taken from Snow’s 1927 magazine article, gives a different figure, it is noted.

The first winter, 1924 to 1925

Sixty-four women in three departments, seven groups: in relay assembly, 16 making the E type of relay and 7 making the A type; in coil winding, 12 winding induction coils, 5 winding receiver coils, and 10 who kept the old lamps throughout as a control; in inspection, 7 inspecting jack shells and 7 inspecting jack springs. Average age 19.8. The existing lamps were Holophane reflectors, 50-watt frosted in relay assembly, 150-watt in the other two rooms; the treatments were General Electric glassteel diffusers with 100-, 200- and 300-watt lamps. Eight periods, the four conditions run through twice in a different order. Light at the bench is given as artificial light alone and as artificial light plus daylight, in foot-candles. Output in relay assembly was measured as average incentive earnings an hour, because Snow could not tell one relay type from another; in coil winding as coils an hour; in inspection as parts an hour.

• Period 1, November 24 to December 26, 1924. Existing lamps. Relay assembly 2.6 foot-candles, 5.0 with daylight; coil winding 4.0 and 10.0; inspection 2.8 and 6.5. Output: E relays 51.3, A relays 67.9; induction coils 11.7, receiver coils 80.9; shells 665, springs 1,211.
• Period 2, December 29 to January 10, 1925. 100 watts. Relay 11.6 and 12.0; coil 8.6 and 16.0; inspection 6.0 and 10.0. E relays 54.6, A relays 68.8; coils 11.8 and 81.3; shells 659, springs 1,138.
• Period 3, January 12 to 31. 200 watts. Relay 25.7 and 25.5; coil 21 and 29; inspection 14.5 and 17.5. E relays 55.5, A relays 69.0; coils 12.4 and 83.8; shells 692, springs 1,230.
• Period 4, February 2 to 14. 300 watts. Relay 42 and 44; coil 36 and 46; inspection 23 and 30. E relays 54.9, A relays 72.8; coils 12.8 and 83.0; shells 695, springs 1,192. In the first week of this period the cardboard baffles and black hoppers went on the lamps to cut the glare, and the instructors were increased from one in ten or twelve to one in eight.
• Period 5, February 16 to 28. 100 watts. Relay 11.6 and 14.5; coil 8.6 and 21; inspection 6.0 and 11.5. E relays 55.0, A relays 71.6; coils 12.7 and 83.9; shells 705, springs 1,236.
• Period 6, March 2 to 7. Existing lamps. Relay 2.6 and 6.5; coil 4.0 and 10; inspection 2.8 and 6.5. E relays 56.3, A relays 74.5; coils 12.9 and 80.9; shells 698, springs 1,272.
• Period 7, March 9 to 28. 200 watts. Relay 25.7 and 33; coil 21 and 38; inspection 14.5 and 27. E relays 56.5, A relays 74.8; coils 12.8 and 84.3; shells 700, springs 1,307.
• Period 8, March 30 to April 11, 1925. 300 watts. Relay 42 and 56; coil 36 and 58; inspection 23 and 40. E relays 55.7, A relays 72.5; coils 12.7 and 85.0; shells 745, springs 1,269.

In periods 2, 3 and 4 the relay room’s reading for lamps plus daylight is no higher than its reading for the lamps alone, which cannot be; the Cornell authors put it down to the plant’s own generator, whose voltage sagged by day. Snow’s published summary rounded the levels to 3, 6, 14 and 23 foot-candles for inspection, 5, 12, 25 and 44 for relays, and 10, 16, 27 and 46 for coils. Snow concluded that light had no effect. The Cornell analysis found that in the first run through the four conditions, from November to February, output rose with light by 0.039 of a standard deviation per foot-candle, and that in the second run, from February to April, it did not.

The second winter, 1926

February 15 to April 24, 1926. Sixty-four women again, in four groups, of which only two were the lighting test: 16 induction coil winders as a control group, moved to another building so that the groups would not compete, and 16 as a test group. The other 32 ran through a test on supervision and one Snow called psychological, of which no numbers survive. Light is lamps plus daylight.

• Period 1, February 15 to 27. Both groups on the existing 150-watt lamps: control 15.8 foot-candles, test 23.7. Output 12.1 coils an hour in both groups.
• Period 2, March 1 to 27. Test group to 200 watts: control 20.5, test 46.3. Output 13.7 and 13.8.
• Period 3, March 29 to April 24. Test group to 300 watts: control 28.1, test 67.2. Output 14.0 and 14.3.

Roethlisberger and Dickson give the test group’s levels as 24, 46 and 70 and the control’s as 16 to 28, and quote Snow: “This test resulted in very appreciable production increases in both groups and of almost identical magnitude.” Snow’s unpublished report credited the rise to supervision, which he said was “of such magnitude as to completely mask any effect of illumination on production.” The Cornell analysis found no significant difference between the groups in any period.

The third winter, 1926 to 1927

September 13, 1926, to April 30, 1927. Thirty-two induction coil winders, matched for output, 16 in each of two identical enclosures built at the east and west ends of the coil winding room, with the sawtooth glass roof papered over and the gaps stopped with burlap, so that the only light was the lamps. Five days to settle in before the first period. The control group had 100-watt lamps and 11 foot-candles throughout. Output is coils an hour, test group first.

• Period 1, September 13 to October 2. Test 100 watts, 11 foot-candles. 13.2 and 12.6.
• Period 2, October 4 to 23. 100 watts, 9. 13.4 and 13.0.
• Period 3, October 25 to November 13. 75 watts, 7. 13.3 and 12.7.
• Period 4, November 15 to December 4. 75 watts, 6. 13.6 and 13.0.
• Period 5, December 6 to 24. 60 watts, 5. 13.6 and 13.1.
• Period 6, December 27 to January 15, 1927. 50 watts, 4. 13.7 and 13.1.
• Period 7A, January 17 to February 7. 40 watts, 2.7. 14.0 and 13.2.
• Period 7B, February 8 to 26. 40 watts, 2.7. 13.8 and 13.1.
• Period 8, Monday, February 28. 25 watts, 1.4. 13.4 and 12.3. One day; the company would not allow the level for longer.
• Period 9, March 1 to 19. Both at 100 watts and 11 foot-candles; the groups swapped enclosures; supervision increased. 13.7 and 13.5.
• Period 10, March 21 to April 9. Both at 11. 14.1 and 13.6.
• Period 11, April 11 to 30. Test 50 watts, 4. 14.1 and 13.8.

Roethlisberger and Dickson’s summary has the test group going from 10 foot-candles down to 3 in steps of one, and the protest coming at 3. Snow’s draft said the test “has demonstrated that the operatives will maintain their productive output at that level or in the neighborhood of that level which is customary, even in the face of insufficient illumination and the attending discomfort.” The Cornell analysis found a significant negative correlation: the further the test group’s light fell below the control group’s, the further its output rose above it. The authors offered less glare as a possible reason and said the reason was unclear.

Hibarger’s test

Homer Hibarger’s “special illumination test” with two relay assemblers is dated by Charles Wrege’s 1976 reconstruction, which François Geoffroy reprints, from January 28 to April 23, 1927: a week at the bench in the department, a move to the dark room on February 4, then 1.0 foot-candle in the morning and 0.39 in the afternoon on February 5, 0.2 on February 6, 0.102 in the afternoon of February 7, then 0.08, then 0.06, then 1.4 on February 28, then 11 from March 1 with the rooms swapped, and from March 21 the supervisor’s desk in the room. Wrege’s table has output rising at 0.2, steady at 0.08, dropping sharply at 0.06 with the women objecting, and settling at 119.5 per cent of the starting figure for the test pair and 117 for the control pair once the supervisor sat among them. Hibarger’s own report, which the Cornell authors quote, says the test “was successful, demonstrating as it did the possibility of maintaining a uniform output under greatly reduced illumination intensities.” Roethlisberger and Dickson place the test after the three winters rather than during the last of them, and say that at 0.06 of a foot-candle “the girls maintained their efficiency.”

What Levitt and List had

Steven Levitt and John List worked not from Snow’s tables but from the company’s own graphs of daily output and light for three rooms, one found on microfilm at the University of Wisconsin at Milwaukee and two at the Baker Library at Harvard, and read the points off the curves. Their timeline has the first wave from November 1924 to April 1925 in all three rooms, light recording suspended through the summer with output still reported, no data at all for the autumn of 1925, lighting experiments in one room only from February 1926, output for the other two rooms until the end of that summer, the blacked-out enclosures in the third winter, the end in April 1927, and output in the first room recorded to October 1927. Their baseline was 4 foot-candles of artificial light, their first-year range up to 36, and their last descent from 11.5 to 1.4. Against the plant as a whole, whose output per worker they estimated from the company’s annual reports to have grown 1.4 per cent a year, the test rooms rose about 10 per cent in the first winter and about 20 per cent in the second. The Cornell authors note that this account leaves out the inspection department, the four-hour Saturdays, the control group of the first winter and the baffles of February 1925.