The different stages of the survey ensure that the process meets the requirements for statistical production:
- tests carried out upstream of the collection, with 53 companies.
- stratified sampling
- reminders and controls of data being collected
- data cleaning
- correction of overall non-response
- calibration
- correction of partial non-response
- winsorizing

As this is a survey of very small enterprises, the response rate is lower than that usually obtained with ICT surveys of enterprises with 10 or more persons employed - it stands at 66.9%, i.e. almost 15 points less than for the ICT 2016 survey. This results in a greater impact of the re-weighting phases in any adjustment of the survey, and therefore in the estimated results, particularly for the accommodation and food services activities, where the response rate is 60.2%.

Before it is used, it is also a good idea to assess the quality of the variables through their partial non-response rate and the confidence intervals provided in the survey quality indicators.

Variables to be analysed with caution (non exhaustive list):
on the type of contract for fixed connections:
CONNEX_FIXE_PRO
CONNEX_FIXE_GP
on the type of contract for mobile connections:
CONNEX_MOB_PRO
CONNEX_MOB_GP
on obstacles to Internet access:
OB_NET_ACCES
OB_NET_COUT
OB_NET_SECU
OB_NET_AUT
on web sales:
BTOC / BTOB
VENT_SITE_ENT
VENT_PL_MARCHE
on the method of transmission used for electronic payment:
TERM_RTC
TERM_IP
TERM_GPRS
on the obstacles to computer use:
OB_ORD_AUT.

These are the variables with the highest partial non-response rates (threshold set arbitrarily at 10%). These variables are to be interpreted with caution as a not insignificant part of the responses has been imputed.

This list is not exhaustive, as other elements need to be taken into account, such as the precision of the estimates and the fact of whether or not the variable is crossed with broadcast domains. Thus, for example, for the qualitative variables, RES_LOC, SPEC_INT, APP_PORTABLE, OB_VW_PROD, as well as most of the variables of module C on computer security are to be interpreted with caution for the sectors J_s, M__ and N_L when SECTOR_8 is crossed with SIZE_2: high standard deviations result in quite wide confidence intervals (more than 10 points).
Among the quantitative variables, the amount of the web sales (VENT_WEB_VAL) also presents problems of precision, with coefficients of variation higher than 20% on most of the broadcast domains. The use of this variable is not recommended and the broadcast of volumes calculated based on this variable is proscribed.

Broadcast domains to be analysed with caution (non exhaustive list):
The Sector of the ICTs (ICT=1), and the crossovers SECTOR_8 x SIZE_2 and ICT x SIZE_2 are to be analysed with caution, in particular for sectors:
J_s: Information and communication; computer repair;
M__: Professional, scientific and technical activities;
N_L: Administrative and support service activities; real-estate activities;
ICT=1: Sector of ICTs.

Crossed with SIZE_2, these sectors of activity present several times with standard deviations greater than 4 points and coefficients of variation that are particularly high for VENT_WEB_VAL.