Lengthy surveys take more time and induce the use of undesired response styles. Response styles generate biases in responses, particularly in lengthy surveys (Baumgartner & Steenkamp, 2001). Response style is defined as tendency to systematically respond to questionnaire items regardless of item content (Baumgartner & Steenkamp, 2001). In practice, market research agencies prefer to use split questionnaires where questionnaires split into parts and each subset is administered to a randomly selected group of respondents (Raghunathan & Grizzle 1995). So far, split questionnaire design literature has been mainly investigated estimation and inference (Lord, 1962; Shoemaker, 1973; Merkouris, 2014; Chipperfield, Barr, & Steel, 2017), data imputation (Raghunathan & Grizzle, 1995; Rässler et al. 2002), and optimal designs (Thomas et al. 2006; Adıgüzel & Wedel 2008; Gonzalez & Eltinge, 2008; Chipperfield & Steel, 2009; Chipperfield & Steel, 2011). Despite split questionnaire is recommended to reduce undesired response styles, its relationship with response styles remains unclear. Two methods of optimal split questionnaire designs were proposed (Adıgüzel & Wedel, 2008): 1) Between-block design: To select entire blocks of questions, 2) Within-block design: To select sets of questions in each block. In this paper, we examine empirically how split designs lessen different type of response styles. Our findings indicate that split questionnaires reduce acquiescence balance, and disacquiescence relative to full questionnaires. The within-block split surveys reduce disacquiescence, acquiescence and balance, whereas the between-block ones did not. Stylistic responding is also influenced by questionnaire satisfaction, attitude, length and mood. Disacquiescence is negatively correlated to mood, questionnaire length and attitude. Midpoint is positively correlated to length, but negatively to questionnaire satisfaction, while acquiescence is opposite for both.

Questionnaires are an important source of data in marketing research. Unfortunately, survey data is often confounded by response styles such as acquiescence response style, disacquiescence response style, extreme response style and midpoint response style. Researchers can use different rating scale formats, which basically differ on two major dimensions, namely Polarity (unipolar versus bipolar) and Anchoring (only positive numbers or negative and positive numbers). To investigate which scale format performs best in terms of minimizing different response styles, we set up an experiment in which we manipulate Polarity and Anchoring. An online survey (N=337) shows strong effects of Polarity and Anchoring on response distributions and provides evidence for the superiority of the unipolar scale format with positive anchors.