Range and Consistency Checks
Most programs for computer assisted data collection have the possibility to specify range checks. The program Interv (De Pijper and Saris 1986) had the facility to specify a lower bound (c1) and an upper bound (c2) for contimuous responses. If the program detects an answer outside the acceptables range , it automatically mentions at the bottom of the screen that the answer should be between c1 and c2 and the responce is ignored which means that the respondent has to answer again.
Hard checks and soft checks
The range checks are hard checks in the sense that the respondent can´t go on if the answer is not given within the acceptable range. It will be clear that one should be very careful with such hard checks in order to avoid problems. Therefore they can only be used to detect very clear typing errors.
Soft checks are available in most interview programs. In such checks conditions are specified under which an extra question will be presented. Such a procedure can be used for consistency checks.
For example a common question is: What did you earn last month?
It is difficult to formulate a range check for such a question because earnings can be as small as 500 euros and as big as 200.000 euros or more. However, most incomes will be between some values, say 1000 and 5000 euros. Given these lower and higher bounds, an answer outside this range is rather unlikely and so the interview can continue as follows:
You indicated a rather unlikely income of .....
If you made a mistake press F1 and answer the question again.
If the answer is correct go ahead to the next question.
This is not a hard check , the answer can be correct even though very deviant and in that case the respondent can go on, but in 90% of the cases the answer will be wrong and the respondent has the chance to correct the error.
Dynamic soft checks
This soft check is already better than the hard check but the procedure will still be rather annoying for people who have indeed an income which is outside this range. Each time they are confronted with this check they may be thinking that the researchers should know better. This is indeed also true. Therefore it would be much better to use this prior information in the interview. This requires that one can store the information in a variable and the variable in a file which can be used in the next interview for the check, We will call this a dynamic soft check. The condition for an extra question would now be as follows:
if (.9*income(t-1)> income(t) or (1.1*income (t-1)) <income (t))
If this condition is satisfied follows:
You indicated a rather unlikely income of ......
while last time your income was .................
If you made a mistake press F1 and answer the question again.
If the answer is correct go ahead to the next question.
In this case the check question is only asked when the new income deviates more than 10% from the last mentioned income. This seemed to us to be a rather efficient way to check on errors without annoying the respondents too much.
Adapting dynamic range checks
In expenditure surveys, on the other hand, one may wonder whether this is the optimal solution. For example one may wonder whether the last price is always the price to store for the next interview. It may be that the price this time was very exceptional. For example when an extra expensive bottle of wine is bought for a special occasion. This can be solved by a simple question whether the deviant price was due to a special occasion. If that is the case then one better does not change the price of the bottle of wine but keep it as it was before.
Another idea that was developed was that it is not necessarily efficient to keep the ranges for all products and at all time the same at a certain percentage like 10% as we saw above. Products vary in price and in speed with which the prices vary. One can take this into account by adapting the ranges on the basis of previous observations. The suggested procedure Kersten et al. 1986) was formulated by adjusting the acceptance interval for the next measure as follows
a(t+1) = 1/2{a(t) +|(p(t) - p(t-1)|}
where a(i) is the acceptance interval at time i
and p(i) is the price provided at time i
If a(t)= 10 and the difference in prices was 20 then the next acceptance interval will increase to 15 while if the difference in price was at that moment only 5 the acceptance interval will become 7.5.
In general one can say if the product varies much in price the acceptance interval remains large but if the difference in price is rather small the acceptance interval will rapidly go down, maybe even too fast so that it gets too small but then one can also determine a minimal level.
In this way one will get by an adaptation process different dynamic range checks for different products and one can guarantee optimal control on the quality of the responses without annoying the respondents too much.
Proactive or Nonproactive Dependent Interviewing
It will be clear that these checks except the hard check are all dynamic range checks which don´t provide the responsents with their previously given answers. Therefore these procedures would belong to the class of nonproactive dependent interviewing methods. In proactive dependent interviews the respondents are provided with their previous answer.
In an interesting paper Eggs and Jäckle (2015) mention that some people accept in a proactive dependent interviewing experiment even information that is wrong but the researchers have not a clear explanation for this behaviour except that it is of course easier to accept a given value than thinking of an alternative, especially if it would take time to remember this information. This would suggest that the nonproactive methods suggested in this section are better.
On the other hand Verwey at al.(1989) could show by asking the same people repeatedly the same monthly income questions in different forms that the large changes in income from month to month obtained by nonproactive methods were not due to errors because also under condition of proactive dependent interviewing the changes in income remained very large. This suggest that most people are willing to indicate changes if the procedure has been made efficient for them.
References:
De Pijper M.W. and W.E.Sais (1986) The formulation of interviews using the program INTERV . Amsterdam , SRF
Saris W.E. (1992) Computer assisted interviewing. Qualitative applications in the social sciences 80, Sage
Prastacos P. , M.M. Recoder and W.E.Saris ( 1992) CASIP : A complete automatic system for information processing in family budget research. The final report of the EUROSTAT DOSES report. Amsterdam, University of Amsterdam
Sala E., Noah Uhrig S.C., Lyn P. (2009) It is time computers do clever things: The impact of dependent interviewing on the interviewer burden. Colchester, ISER.
Eggs J. and Jäckle A. (2015) Dependent interviewing and sub-optimal responding. Survey Research Methods, 15- 29.
M. Verwey, W.E. Saris, K. Mosselman, L. v. Doorn (1989) Tele-interviewing in practice: household incomes show large monthly fluctuations. In: Marketing and Research Today, vol. 17, 230-238
The range checks are hard checks in the sense that the respondent can´t go on if the answer is not given within the acceptable range. It will be clear that one should be very careful with such hard checks in order to avoid problems. Therefore they can only be used to detect very clear typing errors.
Soft checks are available in most interview programs. In such checks conditions are specified under which an extra question will be presented. Such a procedure can be used for consistency checks.
For example a common question is: What did you earn last month?
It is difficult to formulate a range check for such a question because earnings can be as small as 500 euros and as big as 200.000 euros or more. However, most incomes will be between some values, say 1000 and 5000 euros. Given these lower and higher bounds, an answer outside this range is rather unlikely and so the interview can continue as follows:
You indicated a rather unlikely income of .....
If you made a mistake press F1 and answer the question again.
If the answer is correct go ahead to the next question.
This is not a hard check , the answer can be correct even though very deviant and in that case the respondent can go on, but in 90% of the cases the answer will be wrong and the respondent has the chance to correct the error.
Dynamic soft checks
This soft check is already better than the hard check but the procedure will still be rather annoying for people who have indeed an income which is outside this range. Each time they are confronted with this check they may be thinking that the researchers should know better. This is indeed also true. Therefore it would be much better to use this prior information in the interview. This requires that one can store the information in a variable and the variable in a file which can be used in the next interview for the check, We will call this a dynamic soft check. The condition for an extra question would now be as follows:
if (.9*income(t-1)> income(t) or (1.1*income (t-1)) <income (t))
If this condition is satisfied follows:
You indicated a rather unlikely income of ......
while last time your income was .................
If you made a mistake press F1 and answer the question again.
If the answer is correct go ahead to the next question.
In this case the check question is only asked when the new income deviates more than 10% from the last mentioned income. This seemed to us to be a rather efficient way to check on errors without annoying the respondents too much.
Adapting dynamic range checks
In expenditure surveys, on the other hand, one may wonder whether this is the optimal solution. For example one may wonder whether the last price is always the price to store for the next interview. It may be that the price this time was very exceptional. For example when an extra expensive bottle of wine is bought for a special occasion. This can be solved by a simple question whether the deviant price was due to a special occasion. If that is the case then one better does not change the price of the bottle of wine but keep it as it was before.
Another idea that was developed was that it is not necessarily efficient to keep the ranges for all products and at all time the same at a certain percentage like 10% as we saw above. Products vary in price and in speed with which the prices vary. One can take this into account by adapting the ranges on the basis of previous observations. The suggested procedure Kersten et al. 1986) was formulated by adjusting the acceptance interval for the next measure as follows
a(t+1) = 1/2{a(t) +|(p(t) - p(t-1)|}
where a(i) is the acceptance interval at time i
and p(i) is the price provided at time i
If a(t)= 10 and the difference in prices was 20 then the next acceptance interval will increase to 15 while if the difference in price was at that moment only 5 the acceptance interval will become 7.5.
In general one can say if the product varies much in price the acceptance interval remains large but if the difference in price is rather small the acceptance interval will rapidly go down, maybe even too fast so that it gets too small but then one can also determine a minimal level.
In this way one will get by an adaptation process different dynamic range checks for different products and one can guarantee optimal control on the quality of the responses without annoying the respondents too much.
Proactive or Nonproactive Dependent Interviewing
It will be clear that these checks except the hard check are all dynamic range checks which don´t provide the responsents with their previously given answers. Therefore these procedures would belong to the class of nonproactive dependent interviewing methods. In proactive dependent interviews the respondents are provided with their previous answer.
In an interesting paper Eggs and Jäckle (2015) mention that some people accept in a proactive dependent interviewing experiment even information that is wrong but the researchers have not a clear explanation for this behaviour except that it is of course easier to accept a given value than thinking of an alternative, especially if it would take time to remember this information. This would suggest that the nonproactive methods suggested in this section are better.
On the other hand Verwey at al.(1989) could show by asking the same people repeatedly the same monthly income questions in different forms that the large changes in income from month to month obtained by nonproactive methods were not due to errors because also under condition of proactive dependent interviewing the changes in income remained very large. This suggest that most people are willing to indicate changes if the procedure has been made efficient for them.
References:
De Pijper M.W. and W.E.Sais (1986) The formulation of interviews using the program INTERV . Amsterdam , SRF
Saris W.E. (1992) Computer assisted interviewing. Qualitative applications in the social sciences 80, Sage
Prastacos P. , M.M. Recoder and W.E.Saris ( 1992) CASIP : A complete automatic system for information processing in family budget research. The final report of the EUROSTAT DOSES report. Amsterdam, University of Amsterdam
Sala E., Noah Uhrig S.C., Lyn P. (2009) It is time computers do clever things: The impact of dependent interviewing on the interviewer burden. Colchester, ISER.
Eggs J. and Jäckle A. (2015) Dependent interviewing and sub-optimal responding. Survey Research Methods, 15- 29.
M. Verwey, W.E. Saris, K. Mosselman, L. v. Doorn (1989) Tele-interviewing in practice: household incomes show large monthly fluctuations. In: Marketing and Research Today, vol. 17, 230-238
