Analysis Methods
Most of the findings presented in this report are based on tabulations and
cross-tabulations of the survey data. In this section, we describe the methods
used in this work.
Tables
In the descriptive data tabulations of clients
presented in Chapters 5 through 9, the percentage figures in the tables are
based on the total weighted number of usable responses to the client survey,
unless specified otherwise. Responses are weighted to represent clients or
households of all emergency food programs of the A2H network. In general,
weights are based on the inverse probabilities of selection in the sampling and
also account for survey nonresponse. 1 Weights were scaled so that the final
weights represent a month-level count of different clients, as derived in
Chapter 4. 2
Similarly, all tables containing information obtained from the agency survey, as presented in Chapters 10 through 14, are based on the total weighted number of usable responses to the agency survey, unless specified otherwise. The descriptive data tabulations in these chapters represent all emergency food programs in the A2H network. The weights, calculated based on the sampling frame, also reflect survey nonresponse.
Percentage distributions in the client tables are presented by the type of the programs where clients were interviewed (pantries, kitchens, or shelters). When appropriate, the percentage distribution for "all clients" is shown in the last column. Tabulations in the agency tables are presented by the type of programs operated by the agencies.
The percentages in the tables are rounded to one decimal place and are based only on the valid responses. They exclude missing, don't know, refusal, and other responses deemed inappropriate for the question.
The sample sizes presented at the bottom of single-panel tables (or at the bottom of each panel of multipanel tables) reflect the total number of responses to the question. Where the question relates to a subset of the respondents, the appropriate sample size is presented. In general, these sample sizes include missing responses, as well as don't know and refusal responses. The rate of item nonresponse for the client data ranges approximately between 0 and 12% and that for the agency data approximately between 0 and 30% for most variables presented in the tables. We report the percentages of item nonresponse in notes to each table.
The main reason for including only valid responses is to appropriately present the weighted percentage distribution among the main response categories of interest. Our preliminary analysis of item nonresponse revealed little evidence of any systematic biases. Excluding missing data also has the advantage of being consistent with the convention used for two previous studies commissioned by A2H in 1993 and in 1997.
Some tables also present the average (i.e., the mean) or the median values associated with the variable of interest. The average, a measure of central tendency for continuous variables, is calculated as the sum of all valid values in a distribution, divided by the number of valid responses. The median is another measure of central tendency. It is the value that exactly divides an ordered frequency distribution into equal halves. Therefore, 50% of the observations have values smaller than the median and the remaining 50% of the observations have values larger. The median is only suitable for describing central tendency in distributions where the categories of the variable can be ordered, as from lowest to highest.
For selected variables, we have also made annual estimates of the actual numbers of clients (as compared to percentages) falling into various categories. Each of these tables of absolute numbers corresponds to a table with percentages and has the same table number except with the addition of the suffix "N." The "N" tables showing numbers of clients have been computed by multiplying overall annual projections of total clients (as derived in Chapter 4) by the relevant percentages for the client grouping of interest. This must be viewed as only an approximation of the true annual numbers, since, strictly speaking, the percentages have been calculated with monthly weights. However, we believe these approximations to be reasonably good indicators of the true annual numbers.
--------------------------------------------------------------------------------
1. Weights with extremely large values were truncated to reduce variances in the analysis. To keep the sum of weights unchanged, however, weights were then adjusted by an adjustment factor, which is the ratio of the sum of the original weights to the sum of the truncated weights.
2. Weights were originally computed to make the sample representative at the weekly level. They were converted to a monthly scale to take into account the fact that, compared to kitchen and shelter users, a majority of pantry users do not visit the program in any given week.
