re-analysis of the Commission on Community Interrelations experiments

Answering an Antisemitic Remark, 1945 to 1949

A project of the Lab for the Global Study of Antisemitism, Anne Tanenbaum Centre for Jewish Studies, University of Toronto. Ron Levi and Sol Goldberg.

X in

6 publications · 1,130 checks on the transcription · 47 data tables · 3 attitude experiments, 1946 · 7 preference rounds, 1945–1946 · 1 replication, 1949

Harding, Citron and King (1953): transcribed dataset

Companion to CODEBOOK.md (the 1946 anti-Semitic waiting-room experiments).

Levi & Goldberg, transcribed 2026-10-13 from the page images of

Article: Harding, J., A. F. Citron and E. King. 1953. "An Experimental Study of Answers to Anti-Negro Remarks." Journal of Social Psychology 37: 3-17. Received by the journal 4 April 1951. PDF: (held by the authors)

Every row carries source and page. Pages are journal pages (3 to 17); PDF page 1 is a publisher cover, so journal page p is PDF page p - 1. Every number was read from a page image (130 dpi for text, 300 dpi for every table). Values keep the printed form, including the leading + on change scores. The article prints no respondent-level data and no within-cell standard deviations.

Check before use: python3 scripts/verify_hck1953.py recomputes 159 printed totals, ANOVA lines, probabilities, chi-squares and cross-table identities. All pass. Five printed numbers do not reproduce from the cells; they are listed in the script's KNOWN dict and below.

The study

The Commission on Community Interrelations of the American Jewish Congress ran the study on five days, 2 to 6 May 1949, between noon and 2 P.M., at the New York Academy of Chess & Checkers near Times Square. Passers-by were recruited with the 1946 offer of half a dollar for half an hour. The play is set on a crowded subway. A Negro bumps a white man, Bill Jones, who says "These niggers take every chance they get to shove. They're all alike." A third man, Ed Stevenson, answers him with the American tradition argument or the individual differences argument. Every answer is calm (the authors' "low emotion"), and every answer opens with "The train is crowded, and people can't help pushing." In half the presentations a white actor plays Stevenson; in the other half a Negro actor plays him, and the script has him speak as a Negro. Each presentation contains both arguments by the same answerer: version 1 with one argument, a product-identification game, then the incident replayed from Jones's first speech with the other argument. The face sheet at the end asks which of the two methods of answering Jones was best.

The design adds control groups. On 2, 4 and 6 May (12 "pre-and-post-test" groups) the audiences filled out the citizenship scale before the play, after Jones's first speech, after the first argument and at the end, and the about Negroes scale after Jones's first speech in each version. On 3 and 5 May (8 "post-test-only" groups) the first two citizenship administrations and the first about Negroes administration were dropped. The four presentations ran in reverse order on 4 and 5 May. Some 2 May questionnaires were mislabeled and unusable, and 6 May was added to replace them. Replies of Negro subjects were dropped. The 20 groups held 25 to 30 people each, and the summary reports 564 white subjects.

Sign convention

Both scales score high = more anti-Negro. The citizenship score is the number of the statement checked (1 = "best citizens", 9 = "worst citizens"). The about Negroes score is 10 minus the number checked (statement 1, "Jones is ABSOLUTELY RIGHT", scores 9). A change score is the first administration after an argument minus the last one before it, so negative = less prejudice, as in Citron's files (p. 9).

Files

File	Rows	One row per	Contents
`hck1953_cells.csv`	36	printed cell	Table 1 levels, Tables 2 to 4 changes, Table 8 preference percentages, with n
`hck1953_margins.csv`	15	printed total	row, column and grand totals of Tables 2 to 4, for verification
`hck1953_anova.csv`	25	ANOVA line	Tables 5 and 6: mean square ("variance"), df, printed star
`hck1953_tests.csv`	11	reported test	Table 7 t values, probabilities in the text, both chi-squares
`hck1953_design.csv`	33	design item	dates, place, recruitment, sample, program, measures, exclusions
`hck1953_groups.csv`	20	audience group	date, pre-post or post-only, answerer, answer order (reconstructed from pp. 6-7)
`hck1953_answers.csv`	5	answer x answerer	content, manner, opening line, the 1946 counterpart
`hck1953_settings.csv`	3	setting	white answerer, Negro answerer, both
`hck1953_outcomes.csv`	6	outcome	each measure, what lies between its two administrations, the 1946 counterpart
`hck1953_text_numbers.csv`	44	number in the text	page, value, short quote
`hck1953_instruments.csv`	29	scale item	wording and scoring of both scales (appendices B and C) and the preference question
`hck1953_scripts.csv`	31	script line	introduction and the four answer scripts (appendix A, pp. 14-16)

Keys

experiment: hck1953 on every row.

answer: low_american (calm American tradition), low_individual (calm individual differences), both for margins. The ids follow Citron's because the manner matches his "low emotion" condition. The wording is a new anti-Negro script, so the ids name the argument type and manner. They do not mean the same text was spoken.

setting: ts1949_white (white actor), ts1949_negro (Negro actor), ts1949_all (margin).

group_type: pre_post (12 groups), post_only (8 groups), all (20 groups).

answer_position: first or second answer heard. Tables 1 to 3 classify by the first answer, Table 4 by the second.

outcome:

outcome	measure	between the two administrations	1946 counterpart
`citizenship_answer1`	citizenship, 2nd to 3rd	the first argument only	`citizenship_answer1` (Exp III), same position
`about_negroes_change`	about Negroes, 1st to 2nd	first argument; game; Jones repeats his speech	`jones_right` (Exp III, 5 points), same position, 9 points here
`citizenship_answer2`	citizenship, 3rd to 4th (1st to 2nd in post-only groups)	game; Jones repeats; about Negroes scale; second argument	`citizenship_answer2` (Exp III), same position
`citizenship_level_v1`	citizenship level after version 1	(level, Table 1)	none
`about_negroes_level_v2`	about Negroes level after Jones's second-version speech	(level, Table 1)	none
`preference_best`	percent naming each answer best	end of session	the preference question of Selltiz et al. 1950

How the study maps onto the 1946 antisemitism experiments

Answers. Two of Citron's four answers have counterparts: low_american and low_individual. No high-emotion answer was staged. The scripts are new and add a "logic of the situation" line. Content and manner are therefore not crossed here, and Citron's manner-by-content interaction cannot be estimated.
Answerer identity. The second factor is the answerer's race. A Negro actor plays the answerer in half the presentations, and the audience sees him. Citron's Exp III varied identity by a spoken line ("Levy", "I'm Jewish myself") with a different actor. Both studies compare an answerer from the target group with one outside it. The cue differs (seen here, spoken in 1946).
Measures. The citizenship scale has the wording of Citron's Scale III with "Negroes" for "Jews", and the administration points match Exp III. The about Negroes scale has 9 points where Citron's Scale II had 5. Scale I (like Jones) has no counterpart. Means are therefore comparable in sign and position across the two studies; their sizes are in different scale units for the about Negroes measure.
Setting. Both studies recruited near Times Square (the 1946 Exp III also ran two days in Newark). The scene differs (a subway here, an employment-office waiting room in 1946), and so do the date (1949, not 1946) and the room.
Sample. The 1946 study dropped Jewish respondents. This study dropped Negro respondents and kept Jewish ones, who were 46 per cent of subjects (p. 8). The audience here is 97 per cent men.
Control groups. The 1946 study had none. The post-test-only groups here test whether the pretest changes the response; the authors report no difference approaching significance (p. 8).
Unit of assignment. Whole audiences received each condition, as in 1946. Table 5 prints the day-to-day mean square between groups that saw the same version, on 8 or 16 df.

Cautions

Per-group data are not printed. hck1953_groups.csv is reconstructed from the program description (pp. 6-7). The ANOVA df agree with it (12 pre-post groups, 8 post-only, 20 in all).
Bases differ across tables. Table 1 and Tables 2 to 3 have slightly different n for the same cell (90 against 89, 66 against 64), because a change score needs both administrations. The Table 4 margins by answer sum differently from Table 2 because Table 4 classifies by the second answer and includes all 20 groups.
The ANOVA method is not stated. The verifier reproduces every effect mean square in Tables 5 and 6 (within rounding) with weighted main effects and an unweighted-means interaction. Other methods fail on Table 5's 3rd-to-4th interaction (.01), where the weighted residual is negative.
Fisher's combined probabilities (.11, .09) pool the pre-post change and the post-only level for each scale. The verifier reproduces .11 from Tables 5 and 6 with within-groups error terms.
Within-cell SDs are not printed. Table 7's t values allow an approximate SD for each answerer row (SD = |mean| x root n / t), which gives about .8 for both citizenship rows and about 1.3 to 1.5 for the about Negroes rows. The rounding of small means (.01) makes the 3rd-to-4th white SD unrecoverable.
Sample percentages (p. 8) may include the Negro subjects who were later dropped; the text does not say. No table base reaches 564 (the largest, Table 8, sums to 541).

Known discrepancies in the printed article

Printed	Recomputed	Where
p = .09	.079	Fisher combined p for the about Negroes scale (p. 11)
p = .13	.121	t = 1.56 on 146 df (p. 12)
p = .012	.0114	t = 2.55 on 253 df (p. 12)
chi-square 12.2	11.99 (Yates), 13.30 (uncorrected)	Negro-answerer columns of Table 8 (p. 13)
3.50 - .01 = 3.49	3.514	citizenship level from p. 12 against Table 1; bases differ (316 against 313), so this one may not be an error

The first three gaps are small and probably reflect lookup from printed tables. None changes a conclusion the authors drew.

Main results (for orientation)

Content of answer. The American tradition and individual differences answers did not differ on any measure (Table 5 content mean squares .01, .17 and .76 against within-groups .58, 1.86 and .68; Table 6 .00 and .09).
Answerer. On the citizenship scale after the first argument, a white answerer was followed by +.06 (n 166) and a Negro answerer by -.10 (n 147, t = 1.56, p = .13) (Tables 2 and 7, pp. 9, 12). On the about Negroes scale the signs reverse: -.14 (white, t = 1.25) and +.04 (Negro) (Table 3). After the second version the Negro-answerer audiences rose +.14 (t = 2.55, p = .012) and the white-answerer audiences changed -.01 (Table 4, p. 10). Table 5 stars only that color effect (3.06 against .68). Combined across pre-post and post-only groups, the color difference has p = .11 (citizenship) and .09 (about Negroes) (p. 11).
Manner. All answers were calm, so the study has no calm against indignant comparison.
Preference. In all four orders most respondents chose the answer they heard second (80 to 95 per cent, Table 8, p. 13). With a Negro answerer, 95 per cent chose individual differences when it came second and 80 per cent chose American tradition when it came second (chi-square 12.2, 1 df). With a white answerer the figures were 84 and 90 per cent, a gap the authors call not reliable.