Family reconstitution is the technique of linking individual baptism, marriage and burial entries from parish registers back into the families that generated them, so you can measure fertility, mortality and marriage age for a community before modern censuses. You do it by building marriage as the anchor event, attaching each spouse's birth and death, then assigning every child to the right couple, all while tracking exactly when each family entered and left observation. Done carefully it yields rates no aggregate count can; done carelessly it produces confident nonsense. Here is the workflow.

What records and conditions do you need first?

You need continuous registers of baptisms, marriages and burials for one parish over a long, unbroken run, ideally a century or more. Continuity matters more than length: a fifty-year register with no gaps is more usable than a patchy two-hundred-year one. Before you link anything, audit the three series for missing years, illegible stretches and changes of incumbent (handwriting and recording habits shift). Note every gap, because a gap is not zero events, it is unknown events, and it must censor families that span it.

Step 1: Anchor on marriages

The marriage record is your scaffold. Each marriage opens a Family Reconstitution Form (FRF), one record per couple, holding the husband, the wife, the marriage date, and slots for every subsequent event. Capture the verbatim entry and a normalised version:

FRF-0412
  husband: John Aldous, husbandman, of Cratfield
  wife:    Margery Self, of Cratfield
  married: 1671-05-14
  obs_in:  1671-05-14   (marriage)
  obs_out: ?            (to be determined)

The obs_in and obs_out dates are the heart of the method: they bound the period the family is genuinely at risk of producing events.

Step 2: Link births and deaths to each couple

Now attach baptisms of children (matched by both parents' names and plausible spacing) and burials of the spouses and children. Use every qualifier the register gives, occupation, abode, father's name, witnesses, to disambiguate common names.

event	match keys	confidence
baptism	both parents' names, abode, birth spacing	high
spouse burial	name, age if given, abode	medium
child burial	name, parents, age	medium

Record a confidence level on every link. Where you cannot match a name confidently, leave it unlinked rather than guessing, an unlinked event is honest; a wrong link silently corrupts every rate.

Step 3: Set the observation window correctly

A family is only in observation while both spouses are demonstrably present and the wife is within her reproductive span. Set obs_out to the earliest of: a spouse's burial, the wife reaching about 50, or evidence the family left the parish (a child baptised elsewhere, a sudden silence after a run of events). Compute every rate only over the in-observation period.

# years at risk for a fully observed couple
from datetime import date
def years_at_risk(obs_in, obs_out):
    return (obs_out - obs_in).days / 365.25

years_at_risk(date(1671,5,14), date(1689,3,2))  # ~17.8

Why is migration the central pitfall?

A couple that migrates away keeps living and producing events, just not in your register. If you treat the register's silence as "no more births," you fold their later, invisible fertility into a zero and bias the rate downward. The fix is to censor at departure, not at the register's end. When a family's events stop abruptly without a burial, flag them as partly observed and cut obs_out to their last confirmed event, then compute rates only up to that point.

How do you turn linked families into rates?

Aggregate person-years at risk and events within them. Marital fertility, for instance, is births divided by woman-years of married, in-observation, fertile time, conventionally by five-year age groups of the wife:

fertility_rate(age 25-29) = births to wives 25-29
                            / woman-years lived 25-29 in observation

Marriage age comes from the gap between a wife's baptism and her marriage; infant mortality from child burials within twelve months of baptism over births at risk. Each rate is only as sound as the observation windows feeding it.

A practical checklist

1. Audit the three registers for gaps and incumbent changes before linking.
2. Open one FRF per marriage; this is your unit of analysis.
3. Link events with explicit confidence levels; never force a doubtful match.
4. Set obs_in and obs_out to the true presence window.
5. Censor migrant families at their last confirmed event.
6. Compute rates over person-years at risk, by age group.
7. Treat automated linkage as a reviewable first pass, not a final answer.

Key Takeaways

• Anchor reconstitution on marriages, one Family Reconstitution Form per couple.
• Continuity of registers matters more than their total length.
• Link events with explicit confidence; leave doubtful names unlinked rather than guessing.
• The observation window (obs_in to obs_out) governs every rate; set it precisely.
• Censor migrant families at their last confirmed event, not at the register's end.
• Compute rates over person-years at risk, conventionally by five-year age groups.
• Use software for the first pass, but resolve ambiguous links by human judgement.

Frequently Asked Questions

What records do I need for family reconstitution?

At minimum, continuous parish registers of baptisms, marriages and burials for a single parish over a long run of years, ideally a century or more. The completeness and continuity of those three series matter far more than their absolute length.

What makes a family unit observed versus partly observed?

A family is fully observed only when you can date both the start and end of the marriage and account for the wife's entire reproductive span in the parish. If the couple migrated in or out mid-marriage, the unit is partly observed and contributes only to the periods you can see.

How do I deal with common names that match many people?

Use every available qualifier, occupation, abode, father's name, witnesses, to disambiguate, and record a confidence level for each link rather than forcing a single match. Where you cannot resolve a name confidently, leave it unlinked rather than guess.

What is the role of the in-observation and out-of-observation dates?

They define the window during which a family is genuinely at risk of producing the events you count, so rates are computed only over time the couple was demonstrably present. Getting these dates wrong inflates or deflates fertility and mortality estimates.

Can software do family reconstitution automatically?

Tools can propose candidate links and manage the database, but reconstitution still needs human judgement on ambiguous cases, so treat automation as a first pass to be reviewed. Fully automatic linkage on common names produces silent errors that corrupt later analysis.

Why is migration the biggest threat to reconstitution?

Because a couple that leaves the parish stops generating records there, so their later events are invisible and any rate computed as if they stayed is biased. You must censor families at the point they leave observation, not at the register's end.

Related reading

• Nominal record linkage
• Record linkage in history
• Cohort analysis in history
• Standardise occupations
• Clean dates in Python