Statistical processing

Source data

The monthly Labour Force Survey is based on data from the quarterly LFS. The interviews are conducted by telephone and online. In essence, sample data are the same as in the quarterly survey. The answers from the questionnaires are received daily and are checked for errors on a weekly basis in a cadence that is compatible with the monthly calculations. There may occur more thorough examination of the data in the quarterly publication, but this usually only has a negligible effect on the end result. In some months, answers for 5 reference weeks are included, in other months there are only 4 weeks.

Different administrative resources are used to select the sample. Administrative sources are also used to obtain various background information on the people interviewed, for example, on educational level or workplace.

These registers (among others) are being used for the Labour Force Survey: · Central Population Register (CPR) · Population Register · The Register of Labour Market Statistics (RAM) · Register based-labour force statistics (RAS)

Frequency of data collection

Every respondent is interviewed about one specific reference week and the interviews are conducted daily all year round.

Data collection

The monthly Labour Force Survey is supplementary to the quarterly statistics. Every respondent is interviewed about one specific reference week (Monday to Sunday). All questions on work, working hours, unemployment etc. relate to this specific week. Interviews are conducted every day all year. The survey is conducted quarterly and is based on a sample of the population. Data collection - quarterly LFS: The survey is a rotating panel survey including four waves each quarter. Due to the design respondents participate in the survey several times. During one and a half years respondents participate four times. First in two quarters in a row, then an interval of two quarters and then participations in two quarters again. The purpose of the design is to have a theoretical overlap of 50 percent in order to be able to measure both quarterly and yearly changes of employment and unemployment, but the monthly LFS does not have the same overlap and this decreases the accuracy on the monthly estimates.

More documentation on: - Questionnaire (in Danish only).pdf - Dokumentation and Concepts (quarterly). - International definitions(quarterly).

Data validation

Compared to the quarterly LFS, the data validation will typically be of comparable quality. The biggest difference compared to the quarterly validation is that there is a broader check of the data by, among other things, comparing with previous figures and other comparable statistics.

Data compilation

Data is received daily from the data collector and manually checked for obvious errors and omissions before it can proceed for further use. This error correction covers, for example, checks for duplicate interviews, missing labor market status or incomplete interviews. It is limited to the 4 or 5 reference weeks that the individual month contains and it is limited to the age group 15-74. Data is then retrieved from various registers for both the sample and the population, which can subsequently be used to calibrate the weights. The calibration of weights takes place by dividing the sample and the population into a number of groups, based on e.g. gender, age, education and labor market status in the registers in the reference week. A numerical optimization method then corrects the design weights, with regard to the size of the different groups being comparable across the sample and population. This helps to compensate for the biases that arise from a non-random non-response.

The optimal method would be to use the same method as in the quarterly calculation. However, this is not possible, since in a monthly calculation only around 1/3 of the number of observations available for a quarter is available. The design weights are calibrated using various register information to ensure that the weighted sample reflects the population as best as possible. This takes place by forming subgroups for e.g. age and education and ensuring that there is the same proportion in each subgroup in the sample and the population. With fewer observations, one has to make do with less aggregated groupings. In this case, the age groups and register status (provisional labor market status from various registers, i.e. whether you are employed, unemployed or outside the labor market) are more aggregated in the monthly enumeration than in the quarterly enumeration.

The above differences in processing and weighting of data lead to differences in the levels of the enumerated numbers. In order to eliminate structural differences in the levels, the additive Denton-Cholette benchmarking method is used to ensure consistency between the quarterly figures and the average of the monthly figures with the least possible effect on the movements in the monthly time series. The benchmarked time series for unemployment and employment are subsequently seasonally adjusted with models that resemble the respective quarterly counterparts. The totals for unemployment and employment are then used to - in relation to the population - calculate the unemployment rate, employment rate and business rate.

Adjustment

The weighted data is aggregated into total unemployment, employment and population, after which they are benchmarked against the quarterly series. This ensures that the average of the monthly series over a quarter is equal to the quarterly figures and thus no differences arise due to structural differences in, for example, the method for the weighting scheme. The benchmarked figures are seasonally adjusted as far as possible in the same way as the quarterly series, and then the figures are calculation for the unemployment rate, employment rate and employment rate - seasonally adjusted and not seasonally adjusted.

The published figures are calculated based on three totals; the number of unemployed, the number of employed (these two together make up the labor force) and the population - all limited to ages 15-74. There are no seasonal fluctuations in the population as such, so only unemployment and employment are seasonally adjusted. Seasonally adjusted LFS unemployment rate, employment rate and labor force participation rate are then derived.

The two series are adjusted using the X-13-ARIMA method in the JDemetra+ program provided by Eurostat. Since the series start in 2008, which was a very turbulent period in the labor market due to the financial crisis, the models only start in 2010. In addition, manual outliers have been included in the period 2020-2021, due to very large level changes in the series. This ensures that special one-off events do not affect the normal seasonal pattern.