Table 4:
Summary of Findings Depicted in Figure 5 and Robustness to Alternative Follow-up Lengths
| (1) | (2) | (3) | (4) | ||
|---|---|---|---|---|---|
| Years Since Enrollment | N | Untreated Outcome Test Rejects Selection Homogeneity | Always Taker Average Treatment Effect Lower Bound | Local Average Treatment Effect LATE | Test Rejects Treatment Effect Homogeneity (1)*((2)–(3))<0 |
|
| |||||
| Main specification: 20 | 19,505 | −301 [0.003] | 206 (59) | 58 (38) | −44,311 {0.023} |
| 19 | 19,505 | −269 [0.013] | 196 (58) | 52 (37) | −38,565 {0.023} |
| 18 | 19,505 | −311 [0.000] | 210 (56) | 54 (35) | −48,503 {0.010} |
| 17 | 19,505 | −322 [0.000] | 214 (55) | 49 (34) | −52,975 {0.005} |
| 16 | 19,505 | −342 [0.000] | 232 (54) | 56 (32) | −60,245 {0.003} |
| 15 | 19,505 | −381 [0.000] | 211 (50) | 84 (31) | −48,650 {0.015} |
| 14 | 19,505 | −404 [0.000] | 201 (49) | 80 (29) | −49,046 {0.020} |
| 13 | 19,505 | −431 [0.000] | 223 (48) | 75 (28) | −63,808 {0.007} |
| 12 | 19,505 | −443 [0.000] | 191 (44) | 64 (27) | −56,156 {0.010} |
| 11 | 19,505 | −423 [0.000] | 195 (43) | 55 (25) | −59,084 {0.004} |
| 10 | 19,505 | −419 [0.000] | 200 (42) | 47 (23) | −64,017 {0.000} |
| 9 | 19,505 | −413 [0.000] | 192 (40) | 34 (22) | −64,955 {0.000} |
| 8 | 19,505 | −409 [0.000] | 175 (37) | 35 (21) | −57,386 {0.000} |
| 7 | 19,505 | −393 [0.000] | 177 (35) | 46 (18) | −51,740 {0.000} |
| 6 | 19,505 | −412 [0.000] | 185 (33) | 50 (17) | −55,761 {0.000} |
| 5 | 19,505 | −382 [0.000] | 180 (32) | 45 (15) | −51,581 {0.000} |
| 4 | 19,505 | −393 [0.000] | 152 (29) | 46 (13) | −41,568 {0.003} |
| 3 | 19,505 | −354 [0.000] | 104 (23) | 37 (11) | −23,679 {0.012} |
| 2 | 19,505 | −337 [0.000] | 63 (18) | 25 (9) | −12,632 {0.030} |
| 1 | 19,505 | −342 [0.000] | 35 (11) | 20 (6) | −5,194 {0.097} |
Note. Bootstrapped standard errors are under point estimates in parentheses, two-tailed bootstrapped p-values are under test statistics in brackets, and one-tailed bootstrapped p-values are under test statistics in curly braces. Some p-values are zero if the test rejects the null hypothesis in all 1,000 bootstrap replications. The outcome Y is breast cancer incidence, measured at various years since enrollment for all participants, based on initial diagnosis and the exact calendar date of enrollment. The treatment D is mammography, which is equal to one if a participant receives a mammogram in at least one year during the active study period after the enrollment year. The instrument Z is equal to one if a participant is assigned to intervention. The main analysis sample includes women aged 40–49 at enrollment and excludes those who report any breast cancer in their family, any previous breast cancer diagnosis, any other breast disease, or any symptoms, as well as those for whom a nurse found abnormalities or referred them for review. Some differences between statistics might not appear internally consistent because of rounding.