That the nitrate test strips are more precise than the most precise records-based method we examined deserves discussion. Attenuation of odds ratios (ORs) would be substantial even when using the test strip. An observable OR is a function of the true OR and precision, such that when measurement error does not depend on the outcome, the observable OR per unit increase can be estimated by taking the true odds ratio to the power of the square of precision [16, 20]. With a true OR of 2.0, the observable OR when using the nitrate test strip would be 1.43. Use of the best methods examined here for homes and on and off systems (linkage by system and linkage using GIS, respectively) would yield observable ORs of only 1.28 and 1.17. Still, the test strip method relies on subjective comparison of the moistened strip to a colour chart. Bias in ORs due to differential measurement error (including away from the null) might occur if the outcome is already known to the person using the strip [11], whereas linkage-based methods can be applied objectively. Further, for some contaminants, including arsenic [21], a test strip or other in-the-field method suitable for study participants does not exist. Also, it is not always feasible to obtain water from the residence of interest [22], and use of records might allow one to consider past contaminant levels, including those at past residences. The importance of this has been documented [23]. Finally, use of a records-based method might allow a relatively quick and cost-effective study, perhaps without contacting participants. In such case, increasing sample size might be feasible and help compensate for greater measurement error, as well as any added error if one cannot ask participants how much water they consumed, or whether they used bottled water or filters.