Co-twin studies are an elegant and powerful design that allows controlling for the effect of confounding variables, including genetic and a range of environmental factors. There are several approaches to carry out this design. One of the methods commonly used, when contrasting continuous variables, is to calculate difference scores between members of a twin pair on two associated variables, in order to analyse the covariation of such differences. However, information regarding whether and how the different ways of estimating within-pair difference scores may impact the results is scant. This study aimed to compare the results obtained by different methods of data transformation when performing a co-twin study and test how the magnitude of the association changes using each of those approaches. Data was simulated using a direction of causation model and by fixing the effect size of causal path to low, medium, and high values. Within-pair difference scores were calculated as relative scores for diverse within-pair ordering conditions or absolute scores. Pearson's correlations using relative difference scores vary across the established scenarios (how twins were ordered within pairs) and these discrepancies become larger as the within-twin correlation increases. Absolute difference scores tended to produce the lowest correlation in every condition. Our results show that both using absolute difference scores or ordering twins within pairs, may produce an artificial decrease in the magnitude of the studied association, obscuring the ability to detect patterns compatible with causation, which could lead to discrepancies across studies and erroneous conclusions.