Data Collection

Accommodation Survey (January 2013 to September 2019)

Accommodation Survey (January 2013 to September 2019) en-NZ
Accommodation Survey (January 2013 to September 2019) en-NZ

This data collection contains the high level methodology for the current Accommodation Survey from January 2013 to September 2019.




Data source We collect data from accommodation providers or their representatives each month, mostly via a postal survey.

Coverage The Accommodation Survey covers most short-term commercial accommodation in New Zealand.

The target population for this survey is all accommodation providers with the following characteristics:

  • operating on a commercial basis
  • providing mainly short-term (less than one month) accommodation
  • economically significant (generally meaning being GST-registered and having a turnover of at least $30,000 per year)
  • included in class 4400 (accommodation) or class 4520 (pubs, taverns, and bars) in ANZSIC06 (Australian and New Zealand Standard Industrial Classification 2006)
  • classified to 'hotels', 'motels', 'backpacker accommodation', or 'holiday parks'.


  • hosted accommodation (such as ‘bed & breakfast’ establishments)
  • marine vessels (such as cruise ships)
  • private dwellings
  • tramping huts (non-commercial)
  • event-specific accommodation (such as temporary campervan parks)
  • businesses that cease operation or no longer provide short-term commercial accommodation
  • businesses that temporarily shut down (eg for renovations) – we remove them from the survey until they re-open.

Classification of accommodation type We use the New Zealand Accommodation Classification:

  • hotels (including resorts)
  • motels (including motor inns and serviced apartments)
  • backpacker accommodation (including short-stay hostels)
  • holiday parks (including caravan parks and camping grounds).

The predominant capacity provided by a business determines the accommodation type. For instance, if the business provides both motel and camping ground accommodation, but the majority of its stay units are motel rooms, then we classify it as a motel.

Businesses, over time, may change the way they operate, and therefore be reclassified from one accommodation type to another. For example, if a holiday park adds sufficient motel units or backpacker accommodation that it is not primarily operating as a holiday park, then it will be subject to reclassification. This will affect guest nights and other figures for the accommodation types involved.

Interpreting the data

Trend estimates For any series, we can break down the survey estimates into three components: trend, seasonal, and irregular. While seasonally adjusted series have the seasonal component removed, the trend series have both the seasonal and the irregular components removed. Trend estimates reveal the underlying direction of movement in a series, and are likely to indicate turning points more accurately than seasonally adjusted estimates.

We use the X-13-ARIMA-SEATS seasonal adjustment package to calculate the accommodation trend series. The series are based on optimal moving averages of the seasonally adjusted series, with an adjustment for outlying values. The X-13-ARIMA-SEATS package is an updated version of X-12-ARIMA, developed by the U.S. Census Bureau.

The trend estimates towards the end of the series incorporate new data as it becomes available, and can therefore change as we add more observations to the series. Revisions can be particularly large if an observation is treated as an outlier in one month, but we find it to be part of the underlying trend as we add further observations to the series. All trend estimates are subject to revision each month, but normally only the last two or three estimates are likely to be substantially altered.

Differences between trend estimates and month-on-month comparisons Trend estimates reveal the underlying direction of the movement in a series. In contrast, comparisons between one month and the same month in the previous year(s) do not take account of data recorded in between these periods, and are subject to one-off fluctuations. Reasons for fluctuations include changes in the timing of holidays, international crises, and large sporting and cultural events.

Seasonally adjusted estimates We use the X-13-ARIMA-SEATS package to produce seasonally adjusted estimates. Seasonal adjustment aims to eliminate the impact of regular seasonal events. These may be due to climatic effects (such as more guests staying in camping grounds during the summer) or calendar effects (such as holidays). This makes the data for adjacent months more comparable. All seasonally adjusted figures are subject to revision each month.

Moving Holiday adjustment in seasonally adjusted series

From 10 May 2018, seasonal adjustment was improved to better account for moving holidays such as Easter. Corresponding trend series for updated seasonally adjusted series have also changed slightly, reflecting the improved seasonal adjustment.

Easter Most seasonally adjusted series now include an adjustment for the timing of Easter. This accounts for when Easter moves between March and April. Domestic and international guest nights series were not able to be adjusted, due to insufficient data to reliably identify and adjust for the effect of Easter.

Spring school holidays Total monthly guest nights now include an adjustment for the spring school holidays. This accounts for the spread of the spring school holidays between September and October. Only the seasonally adjusted monthly total guest night series is affected by this change. The holiday effects were not strong enough, or predictable enough, to adjust for in other series, such as guest nights by accommodation type.

Introducing moving holiday adjustments also helps the seasonal adjustment algorithm to better calculate the seasonal pattern for other parts of the year. This means that seasonally adjusted movements throughout the entire time series have been affected by the new methodology.

Holidays not directly adjusted for Several holidays that affect guest night levels were considered but not adjusted for due to a variety of factors. These include Chinese New year, winter school holidays, summer school holidays and autumn school holidays.

Sampling Procedure
Response rates by accommodation type for September 2019
Accommodation Type Survey response rate Proportion of guest-night estimate from actual data Proportion of origin of guest estimate from actual data
Hotels 81 92 80
Motels 80 81 66
Backpackers 77 77 67
Holiday parks 76 84 80
Total 79 85 74
Source: StatsNZ

Census The Accommodation Survey is a census, rather than a sample survey. Geographic units (GEOs) on the Business Frame (BF) are identified as potentially in-scope from their industry code (currently a combination of ANZSIC & NZSIC codes). These potentially in-scope units are sent a scoping questionnaire. The scoping questionnaire identifies whether they are in scope (about 90% are), and classifies them with a 2-digit accommodation type. Non-respondents to the scoping questionnaire (about 2%) are not imputed for.

Imputation Imputation Cells Currently, version 2 of the imputation cells are being used. There are 45 imputation cells, based on region and accommodation type.

Imputation Methods This method is applied to the following non-responding units: GEOs that are births and therefore have no previous value GEOs that were tco (temporarily ceased operations) prior to the current month, and did not respond in the current month GEOs that were imputed using this method in the previous month Example: If the total stay nights for the imputation cell is 3000, and the total stay units for the imputation cell is 300, then the cell average stay nights per stay units is 10. If the non-respondent's stay units (actual or imputed) are 12, then the imputed value will be 12*10 = 120 stay nights. (Stay units is always available for non-respondents, as it is part of the scoping questionnaire).

Cell average per guest nights (Mean imputation) Used for:

  • origin of guest variables, but these are imputed as a whole distribution, rather than variable by variable.

The method is as for above, but per guest nights rather than per stay unit. As imputation for guest nights happens first in the imputation system, it doesn't matter that many non-respondents to origin of guests are also non-respondents to guest nights. Example: For simplicity we'll assume there are only 3 place of residence categories, 3 respondents in the cell, and 1 non-respondent

Cell average per stay units (Mean imputation) Used for:

  • stay nights
  • guest nights
  • first nights

So we use the respondents to calculate the 'average distribution' of guestnights across places of residence. i.e. - for place 1 we calculate (10 + 30 + 3)/(17 + 90 + 18) = 43/125 = 0.344 or 34.4%. Now this distribution is applied to the guestnights of the non-respondent, so 0.344 * 40 = 13.76.

Note: Cell average per guest nights imputation was also used for gross income, when this used to be collected in the Accommodation Survey. This imputation method was used regardless of whether the previous gross income data was actual or imputed.

Survey errors The survey aims for 100 percent coverage of the accommodation businesses in New Zealand (a full census). However, in practice, the overall response rate is usually between 76 and 80 percent. We estimate values for the remaining units based on the characteristics of similar establishments in the same or similar regions. This introduces unknown errors into the estimates, and users of the data should bear this in mind. The size of these unknown errors is difficult to quantify.

Other errors include respondent error, and errors in coverage, classification, and processing. Our editing processes identify and remove many errors, but some will likely remain. We cannot quantify the effect of the remaining errors.



1996 -



View Full History
Revision Date Responsibility Rationale
75 30/11/2021 3:29:00 PM