What You Will End Up With
A matrix of existing information about your community, sorted by issue domain: education, employment, housing, health and civic engagement. Under each domain you will have demographic data, a note on where the need is concentrated, and trends from the last three years. Gaps are written down, not hidden. That list of gaps is what justifies, and shapes, any new data collection you do afterward.
Before You Start
Secondary data is information that someone other than you collected. Primary data is what you gather yourself. Use the first before paying for the second.
- Pick the domains and the indicators you want to learn about, so you search for specific things and do not browse.
- Write a source list ranked by perceived quality and ease of access, with the best and easiest to find at the top.
- Find out whether baseline data exists. Baseline data comes from the period before an organization began its work, and it shows the size of the problem at the start.
- Decide who will keep the matrix and where it lives, so the next person can find it and add to it. Existing data goes stale, and a matrix nobody owns goes stale faster.
- Be honest about what you hope to learn without talking to anyone. Secondary analysis can give you demographic data, economic reports and health statistics without direct contact with the community, but it cannot tell you what residents think.
Steps
-
Collect the types of source that usually exist. Look for census data, town records and meeting minutes. Add vital statistics, hospital records, morbidity and mortality reports, and published literature reviews. Hospital admission and exit records can show specific health outcomes, such as adolescent fertility and causes of death. School districts report graduation rates, test scores and truancy for individual schools. Police records hold crime rates and the incidence of problems like domestic violence and motor vehicle accidents. Ask for totals and summaries, never individual patient or case files. Note that a census is an official count taken at set intervals for research, which is different from administrative data collected continuously by an agency.
-
Pull the demographic data. Common categories are race or ethnicity, income level, education level, voter registration status, age and health insurance. Choose the ones that match your issue.
-
Pull the event-based data. Typical examples are traffic accidents, violent crime incidents, disease cases, substandard housing locations, pest infestations and polluted groundwater.
-
Sort everything by indicator and by domain. Each piece of data goes under the specific indicator it measures, and each indicator goes under the issue domain it informs. For every domain, keep three things: the demographic picture, where the need is geographically, and the trend over the last three years, as far back as the records go.
-
Add a map if place matters, and keep it simple. GIS links data to location. A map can show physical features such as towns and roads and political features such as boundaries and protected areas. It needs two kinds of data: spatial data, which locates areas and objects, and attribute data, which describes them. The location data has to exist already and be accurate. If gathering, entering or paying for the data strains your resources, leave things off the map rather than trying to include everything.
-
Record matches and gaps in a matrix. Where outside statistics exactly match your own agency's figures, set up a second small matrix with one row per statistic and one column per source, so you can see at a glance which sources agree. Do the same where numbers disagree, and write down what is missing.
A simple layout for the main matrix:
Domain Indicator Source Trend, last three years Gap or disagreement Education [e.g. truancy rate by school] [e.g. school district] [fill in] [e.g. no data for out-of-school youth] Health [e.g. causes of death] [e.g. hospital records] [fill in] [e.g. clinic data missing] -
Check surprising changes with residents. If several years of data show a notable change in health, educational attainment or employment, talk to community members, for example in focus groups, to understand why. The numbers show that something changed. They do not show the cause.
-
Check who is missing. Current clients and people at your own distributions are a narrow slice of residents who already use community organizations. They cannot be your only source for unmet need.
When the matrix is filled in, read it as a whole before you plan any fieldwork. Look at each domain and ask three questions: what do we already know, how reliable is it, and what is still missing or contradictory. Only the last answer should drive new data collection. If a gap is small, a short conversation with a few residents may close it without a new survey. If a gap is large and matters to your decisions, you now have a clear reason to collect, and a clear question to ask.
Using AI Safely
An assistant can help you sort a pile of documents, but it should not see everything.
- Do not paste confidential client records or sensitive administrative data into a general tool. Administrative data may also come in formats that are hard to use, and a model can quietly garble them.
- Check every figure it extracts against the source document. A model can invent a plausible number.
- Ask it to say what population the data covers and how detailed it is. If it cannot say, go back to the source.
- Keep the judgment about causes for the community conversations in step 7.
Common Mistakes
Trusting old or inaccurate data. Existing data can be out of date or wrong, so check the date and the method before you rely on it.
Assuming the data covers your people. Data gathered for another purpose may leave out the population you care about, or lack detail. Administrative data can be hard to get in a standard format.
Reading numbers as explanations. Secondary data may miss emerging issues, and it does not explain why a problem exists. Test what you see with community members.
Mapping everything. A map is only as good as its hardware, software, data and the people operating it. Start small.
Skipping the baseline. Without information from before the work began, you cannot show how large the problem was.