This section describes the statistical procedures used to analyze NAWS data for this report.  Further details on the statistical procedures can be obtained from the NAWS Web site at http://www.doleta.gov/agworker/naws.cfm

NAWS Weighting Procedure

The NAWS sample is drawn with probabilities-proportional-to-size (PPS) and is designed to be self-weighting.  According to the sample design, each worker has, in theory, an equal chance of being selected for an interview on any given day.  Data limitations, however, make this design difficult to achieve in practice.  For example, there is no accurate measure of the number of workers at any given farm for the weeks of data collection.  This and other small deviations from the sampling plan make it necessary to implement post-sampling weights.  A description of the five post-sampling weights (week, region, cycle, year, and season) and how they are used can be obtained from the NAWS Web site (see URL above).

Determining the Confidence Intervals

A confidence interval is an estimated range of values with a given probability of covering the true population value.  This section provides information necessary for calculating confidence intervals associated with the reported figures.

For categorical variables (e.g., gender, ethnicity, legal status), the proportion or percentage of workers falling into any defined category is reported.  The confidence intervals around the reported survey findings are based on a normal approximation to the binomial distribution.  This method implies that, with a 95 percent confidence interval, reported figures vary at most four percentage points from the true value.  Hence, for example, if 75 percent of the crop workers in the sample are reported within a given category, there is 95 percent confidence that between 71 and 79 percent of crop workers in the overall population actually fall within that category.

For continuous variables (e.g., age, years of schooling, wage rate), measures of central tendency, such as averages or medians, are generally presented.  Confidence intervals for the averages of continuous variables are based on standard errors, which provide a measure of variability of an average value obtained through repeated random sampling from the same population.  A small standard error characterizes an average that varies little from sample to sample, while a large standard error indicates greater variance.  Boundaries of a 95 percent confidence interval around any sample average are calculated by respectively adding and subtracting from the average roughly three times the standard error.  For example, for a variable with a reported sample average of 31 and a standard error of 1, we are 95 percent confident that the true population average is no less than 28 or more than 34.

Table A.1 presents the means, standard errors and confidence intervals for the continuous variables in this report.

Table A.1.  Confidence Intervals for Continuous Variables

Table of Contents