Precision calculations are performed on the variables

With 40,000 units in the sample, it was possible to obtain as parameters:

• coefficient of variation of 0.1 for traditional creators;

• coefficients of variation of 0.08 for self-entrepreneurs.

The estimators may contain a proportion of non-random errors linked to a lack of coverage (see below), measurement errors, non-response, processing errors (data entry) and/or modeling errors.

The sample is drawn in September 2014 from business creations in the first half of 2014 reported in SIRENE on the draw date. Business creations that are not reported to SIRENE on this date are not sampled (about 2% of the

total number of creations).

Conversely, some of the units surveyed are taken out of scope when the data collected in the first wave are processed using the revised SIRENE data. This concerns in particular enterprises with a lifetime of less than 30 days.

For the 2014 cohort, this represents about 1,500 units for the "auto-entrepreneurs" component at the end of the second wave and 1,800 for the classic SINE component.

Partial non-response is subject to imputation: correction of the data by cold-deck (on employment from the EPURE source, on turnover from the ESANE source) or hot-deck (imputation of the other responses to the questionnaire by donor/receiver).

Total non-response is adjusted by imputation, in two steps: imputation of activity status (active or ceased) using an activity status model, then imputation using a

non-response model. Successive reweightings would lead to having three sets of weights in the final file, one for each survey, at the end of the three surveys of the cohort. This would pose difficult problems for crossing several variables from different waves.