Participation is not people

A five-year shea programme commits, in Year 1, to 750 women in village savings groups, 750 women in financial resilience training, 750 women in tree planting training and 750 women in parkland management training. Read quickly, that is 3,000 women. Read properly, it could be 750 women attending four things. Both readings come from the same register, and only one of them is a beneficiary count.

This is not a flaw in the programme. The activities are designed to reach the same women, because that is how resilience is built in one household rather than spread thin across four. The difficulty appears later, when a reporting line called women reached needs a single number and the only available source is a stack of attendance sheets.

What the register answers, and what it does not

An attendance sheet is a record of an event. It reliably answers how many seats were filled and, if it is well kept, who filled them that day. It does not answer how many distinct people the programme has reached, because nothing in its structure links a woman in the savings group list to the same woman in the tree planting list. Names repeat, spellings move, and where a few family names are common, matching on names alone produces both false merges and false splits.

So the honest statement about Year 1 is a range: somewhere between 750 and 3,000 women took part. That range is wide enough to make a results framework difficult to use, which is why the question is worth settling before fieldwork rather than after it.

Building the frame first

A participant frame is a list of identified individuals with enough attached information to find them again and to tell them apart. In practice that usually means, for each participant:

• a stable identifier issued once by the programme, rather than a row number that changes when a spreadsheet is sorted;
• community, group and the activities she is enrolled in, updated as enrolment changes;
• two contact routes where phones exist, and a named group leader or neighbour where they do not;
• the date she entered, so that exposure time can be measured rather than assumed.

Built this way, the frame answers the unique-count question directly. It also does a second job that matters more as the years pass: it makes a panel possible. Returning to the same identified women in a later round supports a statement about change in those women. A fresh sample each round supports a statement about two different groups, which is a different and usually weaker claim.

Building it at the start costs a few days of work and a decision about who owns the list. Reconstructing it in year three is a data-matching exercise with an error rate that is hard to report comfortably.

What a reviewer usually asks

When evidence is challenged, the challenge is rarely about the statistics. It is about whether the people counted were the people meant. Three questions come up most often.

Where did the list come from? An implementer-supplied list is a legitimate starting point and a limitation worth declaring. Say when it was supplied and what was done to check it. Verifying a sample of entries in the field, and reporting how many could not be found, is more useful than a confident silence.

Is a participation being read as a person? If the indicator means people, show the deduplication. If it means participations, label it that way in the table and in the narrative, not only in a footnote.

What happened to the units that could not be reached? Replacement without rules tends to select the most reachable respondents, who are often the ones closest to the road and most engaged with the programme. A reserve list drawn at the same time as the main sample, with the reason for each substitution recorded, costs little and answers the question.

None of this is difficult. It is simply much cheaper to decide in month one than to explain in year three.

What to take from this