Appunti di Piano degli esperimenti sul campionamento

Assume that the target population size is the sum of the units included in the frame, NF , and of the

missing units,NF ̄ , that is, N = NF + NF ̄ , with an undercoverage rate of WF ̄ = NF ̄ /N.

In particular, note from equation that when r = 1 the bias is zero, when r > 1 the bias is negative,

while when r < 1 the bias is positive. Therefore, a large gap between the target and the frame

population can signi cantly affect survey results. These maybe correct for the frame population but

incorrect for the target population.

Item non-response and non-self-selection error

The literature de nes an item non-response as a missing answer to one or more questions in the

questionnaire. we associate this error with a sample survey. The missing answer to one or more

questions in the questionnaire in a census survey is called item non-self-selection. As was de ned

for the unit, item non-response typically has the same probability as item non-self-selection. Case ,

called in the literature missing not at random, is the most severe, since the missing response

depends on the non-observed value of the item and creates systematic differences between

respondents and non-respondents. This can generate bias in the calculation of the parameters of

interest. However, partial non-response creates an incomplete data set and can affect the

application of certain analytical techniques. For example, the application of the Rasch model for

the evaluation of customer satisfaction requires the data set to be complete.

Data collection methods

Firms may use various methods for data collection, according to the various types of customers,

the amount of time and the costs involved. In this section we highlight the features and problems

of a CSS with regard to the types of rms providing services, the types of customers, the data

collection methods carried out throughout the surveys and, consequently, the potential errors

connected with them.

The most common data collection methods and the types of errors associated with them

are as follows:

• Face-to-face interviewing

• Computer-assisted telephone interviewing (CATI)

• Computer-assisted web interviewing (CAWI)

• Web surveys (WS)

• Open Surveys

fi fi fl fi fi fi fi fi