Reporting and Processing of Data
During their visits to atlas units, atlasers recorded all observations on the field cards and submitted these at the end of the season to the project coordinator. After the first year, atlasers were requested to report only observations of evidence for breeding of a species that were equal or stronger than what had previously been found in that unit that (as updated annually and printed on the field card). However, many atlasers reported all their observations, and in order to limit the computer data entry effort, the project compiler reviewed each submitted card and highlighted for data entry, in most cases, only those reports that were species previously unreported from a unit or which were "upgrades" (strengthened evidence of breeding as indicated by the hierarchy of codes). The project coordinator then sent the cards to the data entry team at the ONHP. Initially that team used a software package designed by Kit Larsen (CD creator) to facilitate accurate entry of all data. Upon completing each year�s data entry, Eleanor Gaines (the project�s data entry coordinator at ONHP) sent the electronic files to Kit Larsen, who merged them for querying and a brief annual review by the project coordinator and other members of the steering committee.

Data Quality Assurance
The quality of observations reported is an important concern, as it would be with any project involving hundreds of volunteers, some unknown to the project organizers. Although in many cases the atlas data are better than what existed for some areas (no data at all), every reasonable effort has be made to ensure that quality is adequate. Ultimately, it is essential that the colors for "Possible Breeder" and "Probable Breeder" on this CD�s maps are an accurate reflection of the commonly understood meanings of these terms. However, it is unrealistic to review thoughtfully every one of the more than 124,400 records that atlasers have contributed. Therefore, during the course of the project we tried various procedures to selectively "flag" individual reports for review. Some of these quality assurance procedures were intentionally broad in flagging records. As a result, nearly all project participants have been asked at some point to elaborate on some of the records they had submitted.

The most important problems with atlas data have five main causes, in approximate order of frequency: (1) Participants being uncertain about likelihood of breeding by individual birds they report, (2) Participants accidentally marking the wrong species on the field card, (3) Data entry volunteers misreading cards or punching the wrong computer keys when entering data, (4) Participants not being within the actual hexagon or square from which they report an observation, and (5) Participants misidentifying a species. These five types of errors are described as follows.

1. Uncertain breeding status. This issue was discussed in the Breeding Evidence section.

2 & 3. Data recording and data entry errors. We used three independent procedures to identify reports that might fall in this category: (a) Members of the project steering committee reviewed the maps midway through the project and again at the end, and circled occurrences for which they desired elaboration; (b) For each species, a computerized query identified situations where a species was reported from one hexagon but not from any of its neighboring hexagons; records meeting this criterion were identified and flagged; (c) Lists of species found in each hexagon were compared with lists of species predicted to occur there based on 1993 aerial photographs of vegetation and assumptions about vegetation types used by particular species; if a reported species occurrence was "unpredicted," it was flagged. Once a record was flagged by these procedures, we checked it against the original field card to determine if it was the result of inaccurate entering of data into the computer. When such errors were found, we deleted the record entirely from the database. Another type of error ("omission error") concerns the accidental failure of data entry volunteers to enter some data from a field card. It is difficult to estimate exactly how widespread this error is. It was best detected by providing atlasers with updated field cards for the units they had covered the previous year, and asking them to spot omissions, which we subsequently added to the database.

4. Misreporting of location. This may occur when atlasers unintentionally think they are seeing a bird in one atlas unit when actually one is seeing it in another. Locational misreporting is likely to be greatest for data from the squares because of their smaller size, and for data from units with few roads. There is no sure way to tell when locational errors have occurred. Participants vary in their ability to read the detailed maps we provide, and the maps themselves occasionally contain errors. For cost reasons, Global Positioning System (GPS) technology that could alleviate this problem was not used by most participants.

5. Species misidentifications. This is probably the least frequent source of error. There is no way to conclusively prove when a record is a misidentification. However, we have alleviated most effects of locational and identification errors by noting in the database when a species record has been independently duplicated.� Close to 65% of the unique hexagon-species combinations in our database, and 34% of the unique square-species combinations, are duplicates.