The precision indicators (standard deviation, coefficient of variation, 95% confidence interval) take into account errors due to sampling, total non-response correction, calibration and winsorisation, but not the error related to partial non-response correction.

Variables have been defined for precision calculations on EnDD 2016:
The largest geographic market where the company sells goods or services (A1a A1b A1c A1d)
B1, have a yes among the questions: B1a_ECO_CLIENT_CCD B1b_ECO_CLIENT_CCS B1c_ECO_CLIENT_CCE
C1, have a yes among the questions: C1a_SOC_DISCRI_SEX C1b_SOC_DISCRI_ORI C1c_SOC_DISCRI_SEN C1d_SOC_DISCRI_JEU C1e_SOC_DISCRI_HAN C1f_SOC_DISCRI_ORS
C4_2_SOC_GPECMI50 (C4_2)
D1h_ENV_IMPACT_TOX as follows: None or very low, Low and High or very high (D1h1 D1h2 D1h3) ;
- D2_ENV_CERTIFICATE (D2)
D6_ENV_EFFICA (D6)
E1_RSE_CONNU (E1)
E2_RSE_ACTION (E2)
F6_GVN_PPEXTE50 as follows: no, yes and not concerned (F60 F61 F62)

These variables were chosen as follows:
- the "target" diffusion variables;
- supplemented by a few other variables chosen in order to obtain a sample of variables receiving either a high frequency of positive responses, a frequency close to 50%, or low frequencies, and also in order to test both variables concerning the entire scope of the survey or variables asked only of enterprises with 50 or more employees, or variables filtered by the questionnaire.
The objective is to give the user a quick idea of the accuracy of the indicators he or she obtains based on the size of the sample or sub-sample on which he or she is working.

For the qualitative variables, the indicators are as follows:
average (%) = proportion of "yes" in relation to all units (concerned or not)
standard deviation (s) = root (variance)
95% confidence interval (CI) =[mean +/- 1.96 s].
For qualitative variables with more than two modalities, the calculation is done for each mode.

Total non-response is treated by forming a homogeneous response group (GRH) and then by reweighting, companies' weight is divided by their response probability (that of the GRH).
The variance after calibration of a variable is equal to the variance of the residues of the regression of the variable on the calibration variables.

Errors that do not result from sampling are mainly due to total and partial non-response. Measures to reduce these errors include follow-up and controls during collection and the inclusion of non-response in adjustments.

The directory used to draw the sample is the Statistical Unit Register Registration System. Its coverage of companies is rather good, we consider that all company creations are registered, the SIREN number being mandatory for all administrative procedures. However, not all terminations may be properly recorded.

There is no other survey on the same subject in the public statistical service. The integrated approach to business practices in the different dimensions of corporate social responsibility makes it unique. The survey designers ensured that the questions were relevant in these different dimensions, and in particular that they did not overlap with questions already existing elsewhere (e. g. working conditions in the company or waste treatment).

To avoid measurement errors:
 - the questionnaire was tested with 51 companies face-to-face with an interviewer. There were companies from different sectors, 21 were small companies (20 to 49 employees), with the simplified questionnaire, and 30 had 50 or more employees. As a result of these tests, the questionnaires were modified: some questions were clarified or reformulated, others were deleted or filtered;
 - instructions were given to managers: many questions were open to interpretation such as the feeling of carrying out RSE actions. A two-day training session was held, the content of which was a general presentation of the EnDD 2016 survey and the presentation of collection tools and management application;
 - the managers contacted the companies whose answers came out during the collection checks;
 - a clearance was carried out to make the questionnaires consistent.

Follow-up and controls have been implemented to avoid non-response. The survey was collected from June to October 2016. In addition to telephone reminders, a reminder by mail was sent, followed, if necessary, by a formal notice and a statement of non-response.
The file "Effet redressement sur agrégats.pdf" summarizes the deviations on the aggregate from the adjustment (clearance, imputation and correction of total non-response) for the target variables."