In survey research, computer assisted forms of data collection are rapidly replacing paper and pencil methods. At the same time children are becoming important respondents in many large-scale surveys. However, children possess distinctive cognitive and social developmental characteristics, what must be taken into consideration in the research design. There are many problems to be solved when the respondents are children, including problems of language use, literacy and different stages of cognitive development (Scott 2001).

In order to examine data quality collected with computer assisted self administered questionnaires (CASQ) a study among 9 and 10 years old children was conducted. Within the study two hypotheses were tested:

1. “Cognitive skills do have an impact on reliability of data collected by CASQ, but reliability is not lower than in paper and pencil collection mode.”

2. “Computer skills do not have a significant influence on quality of data in CSAQ.”

In order to answer research questions three different databases were used – PIRLS 2001 (Progress in International Reading Literacy Study) (n=3118); TIMSS 2003 (Trends in International Mathematics and Science Study) (n=3126) – both paper and pencil self administered studies, and computer assisted self administered study where several questions from paper and pencil studies were repeated with the intention to compare the results (n=150). All three studies were conducted in a school setting.

Four constructs were analyzed in relation to the following variables: mode of data collection, index of cognitive level, gender and index of computer skills.

The results show that cognitively more sophisticated children give somehow more reliable results, although further analysis showed the differences between groups are not statistically significant.

Reliability analyses were performed to compare data quality of groups of respondents with high and low index of computer use. Although Cronbach’s α is higher for group of respondents who use computers rarely, the difference between two groups is not statistically significant. Missing value analyses also shows, that respondents who use computers more rarely produced less item nonresponse compared to respondents who use computers more often, meaning that children with lower index of computer skills do not produce lower quality data.