Choose units and disaggregated baseline indicators

Do not begin a baseline by counting whatever happens to be easy to collect. Start with the decision: can the system continue to deliver food and water to households when washing or transport is disrupted? That requires measures of function, response capacity and distribution, not a single production number. Boxes arriving at a market indicate potential availability; they do not establish access for a distant neighborhood. Volume of water distributed indicates a utility service; it does not establish water quality or each user's share. Give every indicator five elements: numerator, denominator where it is a rate, unit, observation window and population covered by the record. Without a denominator defined first, it is easy to compare delivery requests with households as if they were the same unit.

Distinguish stocks, flows and rates. Twenty boxes in the market on Sunday morning are a stock at an instant. Thirty boxes supplied daily are a flow over time. A household service rate needs both the number of households meeting the assessment's stated service threshold and the number of eligible households measured over the same period. Do not sum stocks from different days and call the result weekly supply. Eleven late requests out of 38 requests cannot become a percentage of 42 households. One household may make several requests. If the request-to-household link is unknown, keep the request indicator separate and state its limit. Also record whether a delayed request was eventually fulfilled: timeliness and eventual access answer different questions.

Design an indicator register before data collection: function, operational definition, source, update owner, subgroup, frequency, review threshold and likely error. 'Proportion of scheduled washing-site operating days achieved' is actual operating days divided by scheduled days; it is not the same as the share of food boxes delivered. Choose a baseline period that allows a fair later comparison. A peak-demand week cannot silently be compared with a quiet-month average. If records omit unregistered traders, describe that coverage gap instead of calling the result the entire market. FAO's city-region food-system monitoring guidance supports participatory selection of indicators for desired outcomes; its menu must be adapted to the local question rather than treated as a universal threshold.

Test how nonresponse affects the rate before ranking action priorities. If all twenty nonrespondents did access food, access among eligible households would be 60/80=75%. If none did, it would be 40/80=50%. This is an arithmetic bound, not a statistical confidence interval or a substitute for representative follow-up. Nonresponse may be related to lost connectivity or access difficulty, so do not assume it is random. Separating A and B matters because an aggregate can conceal an operational disparity. Equally, do not make a final deprivation ranking from twenty responses in B without triangulation, such as request records and interviews at distribution points.

The service threshold itself is a decision to disclose. 'Obtained food' might mean buying any item, receiving a defined basket or reaching a seller within an acceptable time; these outcomes are not interchangeable. Before surveying, phrase the operational question in terms respondents understand: could the household obtain the quantity it sought during the disruption day, and by which channel? Do not use this answer to judge dietary treatment or nutritional adequacy; that is outside this program and requires other expertise. Record the channel as well, because market availability could improve while transport remains a barrier for some households. The choice of threshold should be tested with the people whose access is being measured, not only with record owners.

Present a baseline number beside its completeness. 'Respondent access 66.7%; 60 of 80 eligible households responded; two-day window; self-reported access; nonrespondent status unknown' supports a cautious committee decision. Showing 66.7% alone creates false precision. Assign an owner to review definitions when collection changes, and preserve the old version for comparison: revised question wording can move the indicator even if service does not change. Limit the register to measures that guide an action. One indicator updated weekly with a specified review response is more useful than ten columns without an owner or trigger. If a data steward later links requests to households, document the linkage method and its coverage before drawing a cross-unit conclusion.

Baseline units matter. One household can place two weekly orders, and a box may serve more than one household. A market's hundred sales and a committee's eighty households are not a coverage ratio. Match time windows: monthly pumping and weekly orders cannot establish balance during a disruption week. Keep each subgroup's own denominator, and label who recorded each cell and what 'delivered' and 'usable' mean. If eligibility changes, start a new series or transparently recode old records. Ask an independent reader to reproduce every rate from its numerator and denominator.