To avoid confusion with survey questions, researchers aim to create easy questions by constraining question characteristics—e.g. the Flesch-Kincaid Reading Level—in such a way to ensure that the majority of respondents will understand the question (Dillman, et. al., 2009). However, even controlling for specific characteristics, questions can still be complex. For example, two questions can receive the same reading level, but one question might be a harder task for the respondent if it asks for a behavioral frequency while the other could ask for the respondent’s sex. With many surveys moving toward self-administered modes, issues of question complexity become more important since the respondent needs to comprehend all questions without the aid of an interviewer. Researchers have examined how question characteristics in self-administered surveys affect specific response quality aspects such as response times (Yan and Tourangeau, 2008). However, more research is needed to understand the extent of the relationship between complexity and response quality. This study uses the internet component of the Gallup Panel to model the relationship between question complexity and response quality. The complexity measure incorporates page level statistics, such as the number of questions, the reading level, and the type of questions. This index is then used in a cross-classified model to understand the relationship between complexity and different aspects of response quality. Since this study uses data from the Gallup Panel web survey, response quality is measured both in terms of substantive answers (e.g., nondifferentiation, “don’t know” responses) and paradata (response latency, number of answer changes). Preliminary results using one wave from the internet component of the Gallup Panel show that more complex pages lead respondents to have more answer changes per page (correlation of 0.14) and have more don’t know answers per page (correlation of 0.11).