The Japan Tourism Agency’s (JTA) Overnight Travel Statistics Survey (宿泊旅行統計調査) is the first official statistic anyone turns to when discussing Japanese lodging demand. What is far less widely known in day-to-day practice is that the figure for any given month is rewritten each time it is republished. For calendar 2025, the preliminary annual figure released on 27 February 2026 put total guest nights at 653.48 million, -0.8% year on year. The final figure released on 6 July 2026 put it at 661.11 million, +0.3% year on year. A “decline” became an “increase” — the sign itself flipped.

This article verifies how the survey is designed, which stage of the figure appears when, and how far the numbers move on revision, working directly from the actual values in the JTA’s published press releases. We matched 13 corresponding months of first and second preliminary figures, seven months of second preliminary versus final figures, and all 47 prefectures’ annual values, and we quantify the size of the revisions. We also set out how official statistics and a property’s own sales data should divide the work between them.

What the Overnight Travel Statistics Survey Is — a general statistical survey under the Statistics Act

The Overnight Travel Statistics Survey is a general statistical survey conducted by the Japan Tourism Agency under the Statistics Act. Its stated purpose is “to grasp the actual state of overnight travel in Japan on a nationwide scale.” The register of survey targets is built by the JTA from the Establishment Frame Database maintained by the Ministry of Internal Affairs and Communications under Article 27 of the Statistics Act, with the JTA’s own corrections applied. Covered establishments are hotels, ryokan, simple lodgings (kan’i shukusho) and company or organization lodging houses that hold an operating licence under the Hotel Business Act.

The first thing to note is that the survey design itself changed with the January 2026 survey month. Sampling was previously stratified by number of employees; it is now stratified by number of guest rooms. The revision was made to improve statistical accuracy, and the JTA continues to carry a note in its press releases stating that year-on-year comparisons — both ratios and differences — may be affected by the change. How the change in stratification basis feeds through to the tabulated values is examined separately in JTA Lodging Survey Switches Stratification from Staff to Room Count: A Data Read.

The questionnaire itself comes in three formats by room size: Form 1 for 1–19 rooms, Form 2 for 20–199 rooms, and Form 3 for 200 rooms and above. Responses are self-reported; the questionnaire reaches each property by post or online, sent by a contractor on the JTA’s behalf. The reporting deadline is the middle of the following month.

What the design change means in practice is that every property with 20 or more rooms is now in the survey. Previously a property of that same size could fall outside the sample if it had few employees; under the current method the outcome is determined mechanically by room count. Small properties (1–19 rooms) remain a sample survey, with a sampling rate set for each prefecture.

The five property-type categories

Results are tabulated by property type. Ryokan are properties with predominantly Japanese-style construction and fittings; hotels are Western-style and are further split three ways into resort hotels (located in leisure or recreation areas, serving mainly tourists), business hotels (serving mainly business travellers) and city hotels (urban properties that are neither resort nor business). Simple lodgings are properties structured so that sleeping space is shared among many people (bunkhouses, mountain huts, capsule hotels and the like), and company/organization lodging houses accommodate a defined group such as employees (corporate retreats, youth hostels and so on). It is worth confirming which category your own property is tabulated under before making any comparison.

First preliminary → second preliminary → final: a three-stage publication cycle

The figure for any given month is published at least three times: first preliminary, second preliminary, and the final figure compiled on an annual basis. The publication timing of each, and how much questionnaire data is in the tabulation at that point, are as follows.

The mechanism behind the revisions is simple: questionnaires keep arriving after the fact. At the first preliminary stage only about a third of responses are in. By the second preliminary they have built to roughly half, and the final figure picks up whatever arrived later still. The lower a month’s response rate, the more of the estimate rests on imputation — and the more room there is for revision.

The chart below plots effective response rates over the past 13 months. At the first preliminary stage the rate swings between 24% and 40%, bottoming at 24.2% for April 2026; at the second preliminary stage it is somewhat steadier at 44.6–50.1%. That said, the second preliminary rate has drifted down from 50.1% for June 2025 to 44.6% for June 2026, and that shift in level is itself part of the context for reading revision sizes.

Revisions measured — a median 1.9% on guest nights, 0.4pt on occupancy

This is the core of the article. For the 13 months where both first and second preliminary figures have been published (June 2025 through June 2026), we matched the same month and the same metric across the two stages.

The conclusion first: guest nights and the room occupancy rate swing in completely different ways under revision. The revision to guest nights has a median absolute value of 1.89% and a maximum of -3.44% for September 2025 (54.99 million → 53.10 million guest nights, a fall of 1.89 million). The room occupancy rate, by contrast, has a median absolute revision of 0.4pt and a maximum of just +1.0pt for June 2026 (56.2% → 57.2%).

There is also a directional bias. Of the 13 months, guest nights were revised downward in nine and upward in only four. The first preliminary reading tends to come in somewhat high. Occupancy revisions, meanwhile, are evenly balanced — six months up, six down, one unchanged — with no systematic bias.

Where does the difference come from? Guest nights are a total volume measured in person-nights, so guests at properties that responded late are simply added on, and shifts in the population estimate feed straight through. The room occupancy rate, by contrast, is rooms sold divided by total room-nights available — numerator and denominator grow together. Unless late-responding properties have occupancy far off the overall average, the ratio’s level moves relatively little.

Second preliminary → final moves in one direction: all seven months revised up

The revision from second preliminary to final has a different character from first-to-second. Matching the second preliminary values for the seven months June through December 2025 against the monthly figures in the 2025 annual final release, guest nights were revised upward in all seven months. The median revision was +1.87%, with a maximum of +2.61% for December 2025.

And the room occupancy rate was revised downward in all of those same seven months. The range is narrow, -0.2pt to -0.5pt, but the direction is perfectly consistent. Guest nights rise while occupancy falls — two apparently contradictory moves happening at once.

This can be read as telling us something about the character of late-responding properties. Those that miss the second preliminary deadline are relatively small and tend to run occupancy below the overall average. Their guest nights are added, so the total rises; but the occupancy denominator (rooms × days) rises at the same time, and the numerator does not keep pace — so the ratio falls.

In the annual figures, the year-on-year sign flipped

How that accumulation played out at the annual level is unmistakable in the 2025 data. The “2025 annual value (final)” published by the JTA on 6 July 2026 explicitly sets out what changed from the preliminary value published on 27 February 2026.

The year-on-year change in guest nights moved from -0.8% to +0.3%. Anyone who framed 2025 in the first half of 2026 as “domestic guest nights fell year on year,” working from the preliminary figure, ends up with the opposite conclusion on a final-figure basis. The revision itself, at +1.17%, is not large — but because the year-on-year change sat close to zero, the sign changed. When stating a near-zero year-on-year figure from preliminary data, that possibility is worth building in. A separate breakdown by property type of what drove the occupancy revision from 61.8% to 61.6% in this final release is covered elsewhere in our coverage of the 2025 final figures.

Prefecture-level revisions scatter from +6.3% to -2.4%

The national +1.17% is not the result of 47 prefectures moving evenly. Matching every prefecture’s guest nights between the preliminary and final 2025 annual values, revisions were dispersed across an 8.6pt range, from +6.25% in Ibaraki to -2.39% in Kyoto.

Of the 47 prefectures, 36 were revised up and 11 down; the median revision was +0.68%, and the median absolute revision 1.23%. The largest upward revisions were Ibaraki (+6.25%), Fukuoka (+5.41%), Okinawa (+4.87%) and Toyama (+4.80%); on the downward side, Kyoto (-2.39%), Saitama (-2.07%) and Kanagawa (-1.53%) stand out.

Here too the meaning of the year-on-year figure shifts. Fukuoka was roughly flat at +0.9% on preliminary data, but +6.4% on final. Ibaraki went from -5.9% to -0.0%. Kyoto moved the other way, from -0.7% to -3.1%. Discuss a prefecture’s demand trend from preliminary data and the conclusion can change once the final figure lands. When the argument is prefecture-level, the safe course is to label it explicitly as provisional until the final release.

The JTA also footnotes a standard error rate by prefecture for foreign guest nights. For June 2026 (second preliminary) it states that the standard error rate exceeded 20% in six prefectures — Tochigi, Gunma, Chiba, Toyama, Fukuoka and Miyazaki — and that care is required. Because these are estimates from a sample survey, prefecture-level monthly foreign guest-night figures deserve particular caution.

Definitions of the main metrics — room occupancy and capacity occupancy are different things

Below are the definitions of the main metrics per the JTA’s “Explanation of Terms.” The room occupancy rate and the capacity occupancy rate in particular sit at very different levels, so it is always worth checking which one is being quoted.

On nationwide June 2026 figures, room occupancy of 57.2% sits 21.8pt above capacity occupancy of 35.4%. A room designed for two people that is sold to one halves the capacity occupancy rate, so although both are called an “occupancy rate,” they measure different phenomena. When overseas media present Japanese occupancy as low, they are sometimes quoting the capacity occupancy rate. Both formulas, the national average level of each, and how to read them against a break-even point are set out separately in our guide to the room occupancy rate and its national average.

The three major metropolitan areas versus the regions

When discussing the regional dispersion of foreign guest nights, the JTA uses a split between the three major metropolitan areas and the regions. The three major metropolitan areas comprise eight prefectures — Saitama, Chiba, Tokyo, Kanagawa, Aichi, Kyoto, Osaka and Hyogo; the regions are the other 39. For June 2026 (second preliminary), foreign guest nights were -16.8% year on year in the three major metropolitan areas and -7.4% in the regions. That Kyoto and Hyogo are not counted as “regions” is easily misread in discussions of attracting visitors outside the big cities.

Official statistics are the “plane,” your own data the “point” — where the level gap comes from

The Overnight Travel Statistics Survey is a “plane” statistic built by estimating a population. A property’s own PMS or OTA sales data, by contrast, are “point” measurements of what actually happened. Try to compare the two on the same footing and the levels will never line up. To see exactly where the gap opens, we put our own estimated occupancy alongside the JTA’s room occupancy rate for the same prefecture and the same month.

The table below covers June 2026 and sets the occupancy rate compiled by MetroEngines Research (OTA-listed-inventory basis, estimate) against the JTA’s room occupancy and capacity occupancy rates (both second preliminary) for eight major prefectures.

Our estimate exceeds the JTA’s room occupancy rate in all eight prefectures, by a median of 21.5pt — a minimum of +17.4pt in Tokyo and a maximum of +32.2pt in Okinawa. The rank ordering does not match closely either (rank correlation 0.33). Tokyo and Fukuoka sit near the top on both, but Okinawa ranks fourth on our estimate and eighth on the JTA’s room occupancy rate.

Three reasons the levels differ

1. The populations are different. The JTA estimates across every property licensed under the Hotel Business Act, folding small properties of 1–19 rooms into the population via the sample. What we track are properties whose activity is visible on OTAs, which excludes ryokan, minshuku and simple lodgings that are not listed there. Because the JTA side includes many small, low-occupancy properties, its average level is pushed down.

2. Our numerator is “sell-down of inventory listed on OTAs.” Our estimate is calculated from how much listed OTA inventory remains unsold. Most properties do not put every room on OTAs; they list what is left after allocating to direct bookings, corporate contracts and travel-agency blocks. If the listed allocation sells out, the estimated occupancy reads high — while the property as a whole still has rooms available. Conversely, the tighter a property keeps its listed allocation, the higher the estimate tends to run.

3. The weighting differs. The JTA’s room occupancy rate is a ratio built from aggregated rooms sold and total room-nights available, which weights large properties heavily. The gap widens most in Okinawa and Hokkaido, plausibly because resort areas hold many small, highly seasonal properties that weigh on the JTA’s denominator.

Frequently Asked Questions

Conclusion

The Overnight Travel Statistics Survey is an irreplaceable statistic: it measures Japanese lodging demand by prefecture and by property type on a single shared yardstick. It is also a statistic with a three-stage publication cycle, whose numbers are updated as late-returning questionnaires are folded in.

Over the past 13 months, the revision from first to second preliminary was a median 1.89% in absolute terms with a maximum of 3.44% for guest nights, and a median 0.4pt with a maximum of 1.0pt for the room occupancy rate. From second preliminary to final, all seven months moved the same way: guest nights up (median +1.87%), room occupancy down (-0.2 to -0.5pt). At prefecture level, revisions to the annual values scatter from +6.25% to -2.39%, and in several prefectures the year-on-year sign changes.

These revisions are a natural consequence of how the statistic is designed. That is precisely why it pays to state which stage of the figure you are using, and not to hang a conclusion on a few-percent move at the preliminary stage. The finding in this article — that the room occupancy rate is more stable than guest nights even at the preliminary stage — doubles as a guide to reading preliminary releases.

And what official statistics measure is the “plane.” The “point” movements — your own booking pace, how competitors are selling — are faster to follow in data you can pull daily. The two are not in competition: check the level with official statistics and follow the movement with daily data, and only together do they give you the resolution a decision needs.

Related Reading

• JTA Lodging Survey Switches Stratification from Staff to Room Count: A Data Read
• June 2026 Occupancy 56.2%: Guest-Nights per Room in 47 Prefectures