The 2002 campaign study was different from other publicly released polls, as it used a 'tracking' or 'rolling cross-section' method. This approach to the studying of elections was pioneered in the United States during the 1980s, and first used comprehensively in the 1988 Canadian Election Study. The 1996 NZES was the second national election study to use the method, followed by the Dutch in 1998, and the British Election Study in 2001.

The sample was based on an average of 100 interviews per night. To get the best impression of campaign trends, the daily data are averaged over five days. Each daily data point shown here represents an average of the day in question and the two days before and after it. This both reduces sampling error and smoothes out short-term movements that have no effect on the overall trend. Thus the N for each data point in the figure is 500. The maximum margin of error possible within 19 out of 20 possible samples is therefore + or -4.4 per cent assuming a 50:50 distribution between two frequencies, and + or -1.9 per cent assuming a 5:95 distribution. Sampling was from telephone numbers supplied by Telecom, and the data is weighted by household size, age, gender, and region (Auckland against the rest of New Zealand). An additional weight adding recall of 1999 party vote was applied over the first 10 days, but made no substantive difference to the estimates, and was therefore discarded.

The campaign study was conducted for the NZES by ACNielsen (N.Z.) Limited.

Any differences between the NZES campaign study and other polls are likely to be due to the 20-minute length of the NZES questionnaire and the positioning of the party vote question after several others asking for opinions on issues and leaders. Respondents are therefore encouraged to consider these matters before indicating their vote intention, whereas in other polls the vote question is likely to precede all others. It is a matter of debate which practice produces the best estimates of voting intention.