Response latencies answering to attitude questions can be used as a measure of chronic attitude accessibility. But depending on the theoretical interest, several determinants of response latencies have to be treated as bias effects and should be statistically controlled to adequately interpret response latencies. Main bias factors are the individual baseline speed of respondents reflecting a constant individual characteristic of mental speed of information processing, effects of the measurement instrument and situational effects. In this study using response latency data of a nation-wide German survey (CATI), four statistical transformation methods to control the individual baseline speed are empirically evaluated and compared: Z-Score, Difference Score, Ratio Index, and Rate-Amount Index. The empirical findings support the assumption of an increased data quality transforming "raw" reaction times into indices controlling the baseline speed. Additionally, the data quality increases if additional systematic bias effects are controlled (here: question order, effect of extremity). 1 INTRODUCTION In attitude theory, response latencies answering to attitude questions are regarded as an indicator of attitude strength. Defining an attitude as the association between an object and its evaluation (Fazio 1986, 1989, 1990b), the strength of an attitude is the strength of this association. Response latencies are often used to measure the chronic accessibility of attitudes. This accessibility points directly to the mental process during the activation of an attitude and is regarded as a measure of the associative strength. Therefore, an attitude is assumed to be stronger if it is easily accessible, measured by a shorter response latency, and to be weaker if it is less or not accessible, measured by a longer response latency. In recent decades, the development of modern techniques of computer assisted interviewing has made it possible to measure response latencies to attitude questions in the context of large scale survey studies (Bassili/Fletcher 1991, Bassili 1993, 1996b). In contrast to the laboratory context, the measurement of mental information processing in a relatively uncontrolled survey context is much more biased. For example, different interviewers may measure the reaction time with different accuracy (raw response latencies implicate the latency of the interviewer to press the appropriate key), or the respondent may be distracted by the presence of others or unforeseen events. Additional problems appear if the respondent fails to answer 'correctly' if he or she has difficulties to understand the question or to generate an answer and translate it into the given categories or scale (see next chapter).