Church Attendance Trends: Telling Seasonality From Decline

A 12-month attendance line falls every summer and spikes at Christmas, and neither movement means anything on its own. To see decline you compare each week against the same week last year, smooth the line with rolling averages, and count households rather than heads.

Key points

• Nationally, week-to-week attendance is flatter than church folklore assumes: cellphone geodata for over 2 million Americans found limited week-to-week seasonality, with Easter and Christmas nearly 50% above a typical week as the major exceptions.
• Christmas is the largest distortion in your year. The Church of England counted 1.96 million people at Christmas Eve and Christmas Day services against an average Sunday attendance of 590,000. It is counted across two days and every service on them, so it is not a like-for-like ratio.
• There is no reliable published benchmark for how big a summer dip should be in one congregation, so your own July last year is the only honest comparison available.
• Rolling 4-week and 13-week averages remove week-to-week noise; same-week-last-year removes the season. You need both, and they answer different questions.
• A flat total can hide real decline, because a congregation where every household moves from three Sundays a month to two produces a third fewer people with nobody leaving.

Every church I’ve worked with has the same argument in August. Attendance is down, someone says it’s just summer, someone else says it was down last September too, and nobody can settle it because the only evidence is a number on a slide and everyone’s memory of last year.

That argument is settleable. It needs two hours and twenty-four months of data.

What does a normal year of church attendance look like?

Flatter than you’d think, at the national level.

Devin Pope’s working paper on religious worship attendance measured from smartphone geodata tracked over 2 million Americans across 2019 and found roughly 45 million people attending a worship service in a typical week. The week-by-week series is close to level. The two real spikes are Easter Sunday and the week of Christmas, both running nearly 50% above a typical week, and the only dips he records are small ones on holiday weekends: Memorial Day, Labor Day and Thanksgiving. Be precise about what that paper is: an NBER working paper, circulated for comment and explicitly not peer-reviewed, measuring where phones went rather than asking people where they went. It is still the best week-by-week measurement of American attendance there is.

No summer trough in that data. Which is a useful thing to know, and not the thing most pastors expect.

The reason is aggregation. A national series adds up hundreds of thousands of congregations across every climate and school calendar in the country, so local seasonality cancels out: the beach town empties in July while the mountain town fills. Your church is not the national aggregate. It is one congregation, in one school district, with one set of vacation habits, and it can have a seasonal shape the national line has no way of showing.

The holiday spikes, though, survive aggregation, because they land on the same dates everywhere. The Church of England publishes the clearest version of this in its annual statistics: 1.96 million people at Christmas Eve and Christmas Day services, 1.03 million at Easter, against an average Sunday attendance of 590,000 and 707,000 across a typical week. Christmas is counted across two days and every service on them, so it isn’t a like-for-like ratio, but the scale is unmistakable. Easter runs around 1.7 times a normal Sunday. Christmas runs over three times it.

Now the honest part. I went looking for a citable national figure for the size of the summer dip in one congregation and could not find one from a source I’d stand behind. The numbers circulating for it trace back to vendor surveys and blog posts, not to Gallup, Pew, Hartford or a journal. There is no benchmark to hold yourself against, so build your own from your own history. A figure drawn from churches in other climates was never going to tell you much about yours.

Why doesn’t your weekly total show decline?

Two pieces of arithmetic, and both are working against you.

Arrivals cancel departures. Eleven families drift out over a quarter, a handful of new ones start coming, and the total holds. Nothing in that number says eleven families are gone, which is the whole reason finding out who stopped coming is a spreadsheet project rather than a report you can run.

Frequency moves independently of membership. This is the bigger one and it’s badly under-appreciated. Barna’s 2025 tracking puts churched American adults at an average of 1.6 times a month, roughly two out of every five weekends. Barna reports that figure alongside rising attendance among younger adults, so it describes rhythm rather than decline. Pope’s cellphone data finds only 5% of Americans attending weekly against the roughly 22% who tell surveys they do. Pew’s 2023–24 Religious Landscape Study puts one-third of US adults in a service in person at least monthly, including 25% at least weekly.

Take that into a single church. Four hundred committed households attending three Sundays out of four produce a very different Sunday count from the same four hundred attending two out of four. Nobody has left. Nobody is on a lapsed list. Your roll is identical and your room is a third emptier.

The Church of England reached the same conclusion and did something about it. Its statistics office added a second measure in 2012 because, as it puts it, many churches told them that decreases in average weekly attendance reflected a reduction in how often people come to church rather than a fall in the number of individuals who are regularly part of the church, so parishes now report a “worshipping community” alongside the attendance count. That means anyone attending regularly, defined as about monthly. Two numbers, because one of them was answering the wrong question. It’s the clearest statement of this problem I’ve found from anyone who counts for a living, and it’s why the metrics worth tracking are mostly rates rather than totals.

A third thing moved under everyone’s feet: where attendance happens. Gallup found in-person attendance at 26% in May 2023 with 5% attending virtually, essentially the reverse of April 2020’s 4% in person and 27% virtual, and Pew finds 23% of Americans watching services online or on TV at least monthly, with 40% participating at least monthly in person or online or both. If your in-room count fell in 2020 and never came back, some of that is loss and some is a household that now watches twice a month and comes twice a month. Check-in data cannot tell those apart.

The wider picture isn’t uniformly grim. Hartford’s Faith Communities Today research reported in April 2026 that median in-person weekly worship attendance rose for the first time in 25 years, to 70, above the pre-pandemic median of 65, across 7,453 congregations. A median of 70 is worth sitting with: most churches are small enough that a dozen people moving from weekly to fortnightly is a visible dent in the line.

Four comparisons that separate seasonality from decline

Each one removes a different kind of noise. Run all four and the arguments in the August staff meeting stop.

Same week last year

The only comparison that removes seasonality outright, because seasonality repeats on the calendar. Week 30 of this year against week 30 of last year has the same school holidays, the same weather regime and the same vacation season built into both sides.

Two things break it. Easter moves anywhere between 22 March and 25 April, so a March-to-March comparison can hold Easter one year and not the next. And a 53-week year shunts everything by one. Align Easter and the weeks around it on the liturgical calendar rather than on week numbers, by hand, once, and leave a note in the sheet saying you did.

Rolling four-week average

Kills the one-off. A snowy Sunday, a stomach bug going round the kids’ ministry, a long weekend, a power cut. None of those are information about your church, and all of them will be read as information if you show a raw weekly line to a board.

Rolling 13-week average

A quarter, smoothed. This is the number to put in a leadership pack, because it moves slowly enough that when it moves, something happened. Thirteen weeks is also a native unit in Planning Center: dashboard metrics widgets offer 30 days, 13 weeks or 12 months, with a year-over-year comparison, so you can eyeball the shape before you export anything.

Households, not heads

A family of five leaving reads as five people. A new couple arriving reads as two. In heads that swap looks like a loss of three; in households it’s one out and one in, and those are very different facts about your church.

The catch is mechanical: the Household ID is not saved to the profile and exists only in the current import, so it isn’t a stable key you can join exports on. Rolling to household means matching on surname and address by hand, which is why almost nobody does it.

What the bumps in your line are

Before you interpret a chart, subtract the artifacts. Most of what looks like a trend in a raw attendance line is a counting effect.

Months with five Sundays. A monthly total is not comparable month to month, because a month has four or five Sundays and the five-Sunday month is 25% bigger for free. Either report an average per Sunday or count the distinct Sundays in every month and divide. This one silently corrupts more board packs than anything else on the list.

Easter’s date. It moves by up to five weeks. Any month-on-month or quarter-on-quarter comparison spanning March and April is comparing two different things half the time.

Christmas Eve cannibalising Sunday. Christmas Eve services usually live in a separate Check-Ins event, so the Sunday chart can dip in the week your building was fullest all year. Read the two together or the busiest week of your year registers as a bad one.

A change in how you count. Moving from clickers to check-in stations, or switching a room onto Headcounts, which records the number of attendees without individual details, will shift your number by more than a year of real drift. So will adding or dropping a service time. Annotate every such date on the chart, because in two years nobody will remember and the step will get explained as a trend.

A breakdown you haven’t checked. Before splitting the line by check-in type or location, read what Check-Ins actually records: a mismatch between Services team names and Check-Ins locations moves people between the Regular and Volunteer buckets without moving anyone in the building.

How do you build a 12-month attendance trend from Planning Center?

Planning Center holds everything you need for the counts. It will not do the comparisons for you, and what each attendance report actually returns is worth knowing before you start clicking. Budget two hours the first time.

1. Open the event chart on your weekend Check-Ins event. It plots the number of unique people checked in or counted in Headcounts over a timeframe you set, and the download icon exports what’s on screen as CSV.
2. Set the timeframe to 24 months, not 12. Every comparison here needs a prior year on the other side of it, and going back for the second year afterwards is the most common wasted step.
3. Check what a row is. You want one row per date. This CSV is counts by date, not people. That distinction costs an afternoon if you assume the wrong one, since a custom report on the Reports tab is what returns names.
4. Add a Sunday index. For each date, its ISO week number, its ISO week-year, and the count of Sundays in its month. The week-year is not the calendar year: 1 January 2023 belongs to 2022-W52, and keying it as 2023-52 collides with 31 December. Sheets has no ISO-year function, so use =YEAR(date-WEEKDAY(date,2)+4), the year of that week’s Thursday. The last column is what stops a five-Sunday month reading as growth.
5. Add rolling averages. A 4-week and a 13-week column beside the raw count. Both are one formula dragged down a column.
6. Add same-week-last-year. Build a key from the year and ISO week columns, then look each row up against the key for the prior year (XLOOKUP on (isoyear-1)&"-"&week) and take the difference as a percentage. Week 53 exists in some ISO years and not the one before, so give the lookup a fallback, IFNA(..., week 52 of the prior year), or that Sunday’s comparison comes back #N/A and the 13-week average beside it drops a point. Don’t shift the series by 52 rows: one cancelled Sunday, one extra midweek event date, or a 53-week year puts every comparison below it one week out, and the error is invisible because the numbers still look plausible. Then fix Easter by hand: find Easter in each year, and align that week and the two either side of it on the liturgical date rather than the week number.
7. Annotate every known event. Easter, Christmas Eve, snow closures, the week you added a service, the week you switched on a new check-in station. A column of notes on the same rows as the numbers.
8. Repeat at household level, once you trust the counts. The event chart cannot do this. You need person-level rows from a custom report, stacked session by session, matched on Person ID, then given a household key by surname and address and reduced to distinct household-and-Sunday pairs before anything is counted, so a family is present once for any Sunday any member came. Slow, and the part that answers the question you actually asked.
9. Write the decision rule down before you look at the finished chart. Mine is three consecutive non-overlapping 13-week averages below the same period last year, read at quarter ends, not week by week, because the weekly rolling column makes three adjacent readings the same fifteen weeks three times over. Any rule beats deciding what counts as decline while staring at a line you already have an opinion about.

If this is going to be a recurring job rather than a one-off, do it in a sheet: getting Planning Center exports into Google Sheets covers the import and the Person ID match, and turns the monthly redo into a paste rather than a rebuild.

How do you read the line once you have it?

Start with one number: the 13-week average against the same 13 weeks last year. Everything else on the chart is context for that.

Then split it, because “attendance is down” is two entirely different problems wearing the same coat.

Households down, frequency flat. Families have left. That’s a pastoral list, not a programming question, and the useful next step is naming them rather than discussing the trend. Nothing built into Planning Center compares a household’s rhythm across two windows, so that list gets built by export and pivot.

Households flat, frequency down. The same people are coming less often. Nobody to call, because nobody has gone. This is the harder problem and usually a question about what a Sunday is for, whether the rhythm of the year has changed, or whether a chunk of your congregation quietly moved to watching online twice a month.

Both down. Check first-time guests before anything else. A church losing households while holding its guest numbers is leaking; one losing households with guest numbers falling too has a front-door problem that started earlier, which is why first-time guest tracking belongs on the same page as the trend line.

Neither down, and you’re still worried. Trust the worry enough to check serving. Volunteers tend to stop being available before they stop attending, and it is the fastest-moving signal you have.

One thing not to do: read a single quarter as a trend. A quarter can be a school calendar shift, a building project, or a run of bad weather. Three consecutive quarters down against the prior year is a fact about your church. One is a reason to keep the sheet open.

And do not reach for a national figure to settle it. Gallup has weekly-or-nearly-weekly attendance among US adults falling from 42% in 2000–2003 to 38% a decade later and 30% in 2021–2023, and 31% in 2025, driven mainly by growth in the share of Americans with no religious affiliation rather than by churchgoers thinning out. That tells you the direction of the sea you’re swimming in and nothing about your boat. A national average is a distribution, and the median congregation of 70 sits inside that same decline. Your own two-year comparison is worth more than every national figure in this post combined, which is the argument for spending the two hours.

The part worth automating

The sheet works. The problem is that it works once. Rebuilding it every quarter falls to the person with the least free time, and a 12-month trend that’s four months stale is a chart about a church that no longer exists.

That’s the specific job Parable does: it reads Planning Center nightly into one place where Check-Ins, Giving, Groups and Services share a person key, so the rolling trend, the same weeks last year, and the households behind the drop are one query rather than three exports. The recurring version lives on a church metrics dashboard, with SQL and warehouse access underneath. How to track church attendance covers the step before this one, if you’re still deciding how to get the numbers in at all.

If your attendance question is “how many came,” Planning Center answers it already and a second login would be a step backwards.

The question this post is about is different. It’s whether the number going down means anything, and that one is never answered by a bigger number. It’s answered by the same number, measured against itself, a year ago.