How to Compare When Data Come from Diverse Sources: A 4-Level Model of Change in Institutional Trust over Time

Much effort has been invested by International Survey Programs to harmonize data a priori. However, different survey programs may take different decisions in this regard. And we may need to combine data coming from these different sources. This paper will show how to use multilevel modelling to analyze data collected on the same topic even when the question wordings, answer scales and other features vary between surveys. We use institutional trust as a “showcase”. Most of the research published on this question has been carried out in Europe and North America in which comparable data are more easily available. Outside these regions, comparing trust between countries is challenging. The Global Barometers, the WVS and LAPOP are the main sources of data and since not much data are available, we need to use all that we have. These data are disparate: The answer scales differ (4 anchors for the Barometers and the WVS, 7 for LAPOP). More importantly, the institutions on which trust is assessed vary both between countries and over time. The paper compares change in institutional trust over 20 years in South and Central America, Africa, West Asia and North Africa, Asia and Eastern Europe. The data combine more than 800 surveys conducted in these countries by various Survey Programs. Instead of trying to harmonize these data a priori, the paper uses a 4-level model which allows to use all the data and control a posteriori for the differences between the surveys. Questions on trust for different institutions (level 1) are analyzed as nested within respondents (level 2), themselves nested within years (level 3) in which the surveys were conducted and within countries (level 4) where people live. The analysis allows for assessing whether the use of different scales has an impact of the level of trust.