On the Practice of Dichotomizing Quantitative Variables

Department of Psychology, Ohio State University, Columbus 43210-1222, USA.
Psychological Methods (Impact Factor: 4.45). 04/2002; 7(1):19-40. DOI: 10.1037/1082-989X.7.1.19
Source: PubMed


The authors examine the practice of dichotomization of quantitative measures, wherein relationships among variables are examined after 1 or more variables have been converted to dichotomous variables by splitting the sample at some point on the scale(s) of measurement. A common form of dichotomization is the median split, where the independent variable is split at the median to form high and low groups, which are then compared with respect to their means on the dependent variable. The consequences of dichotomization for measurement and statistical analyses are illustrated and discussed. The use of dichotomization in practice is described, and justifications that are offered for such usage are examined. The authors present the case that dichotomization is rarely defensible and often will yield misleading results.

Download full-text


Available from: Shaobo Zhang,
  • Source
    • "Given the evidence of dimensionality of SI presented here, this practice, except in a few rare in- stances, 2 is methodologically questionable. Reducing continuous constructs into arbitrary dichotomies results in a corresponding reduction in measurement precision (Ruscio and Ruscio 2002) and statistical power (Cohen 1983; MacCallum et al. 2002), and in some cases may yield spurious statistical findings (Maxwell and Delaney 1993). The treatment of dimensional constructs in this manner may substantially hinder attempts to elucidate the complex network of causal factors underlying the construct of interest. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Despite the strong association between suicidal ideation and suicidal behavior, a relative minority of ideators transition to attempting suicide. Clarifying the latent structure of suicidal ideation has direct implications for theory, as well as suicidal risk assessment and prevention efforts. Taxometric analysis is a statistical technique specifically designed to assess whether a latent construct is taxonic (i.e., categorical) or continuous (i.e., dimensional) in nature. Although this statistical approach has been increasingly used over the past decade to elucidate the latent structure of various forms of psychopathology and related risk factors, there are no taxometric studies to date of suicidal ideation. The aim of the current project is to apply taxometric methods to a sample of clinically depressed, treatment-seeking adolescents (n = 334). Current suicidal ideation was measured using the Suicidal Ideation Questionnaire-Jr., (SIQ-Jr.). The results of two mathematically non-redundant taxometric approaches (i.e., MAXEIG and L-Mode) are consistent with a continuous latent structure for suicidal ideation. The current findings suggest that suicidal ideation in depressed adolescents is dimensional. The implication of these findings for research, theory, and suicidal risk assessment strategies are discussed.
    Journal of Abnormal Child Psychology 11/2015; DOI:10.1007/s10802-015-0022-y · 3.09 Impact Factor
  • Source
    • "Conversely, situations in which a factorial design is suboptimal and comes with the risk of nontrivial loss of information comprise, e.g., research on lexical stimulus properties such as frequency of occurrence or acoustic duration. Although dichotomization into high versus low frequency, or long versus short acoustic duration, makes it possible to apply analysis of variance, dichotomization of continuous variables has many undesirable properties (Cohen, 1983; MacCallum et al., 2002; Baayen, 2010), and should be avoided if only to avoid unnecessary loss of power. For research questions addressing the consequences of such numeric stimulus properties for human processing, regression is the appropriate statistical tool (see also Balota et al., 2004). "
    [Show abstract] [Hide abstract]
    ABSTRACT: Unlike molecules or plots of barley, subjects in psycholinguistic experiments are intelligent beings that depend for their survival on constant adaptation to their environment. This study presents three data sets documenting the presence of adaptive processes in psychological experiments. These adaptive processes leave a statistical footprint in the form of autocorrelations in the residual error associated with by-subject time series of trial-to-trial responses. Generalized additive mixed models (gamms) provide a unified framework within which both factorial predictors and covariates given with the experimental design, as well as non-linear random effects and interactions with experi- mental time can be uncovered and evaluated. Gamms not only provide substantially improved fits to experimental data with time series structure, but also provide improved insight into predictors of theoretical interest, as well as a more refined window on the random effects structure. Our res- ults challenge the standard advocated by Barr et al. (2013). The analytical cage of the maximal linear mixed model to which this standard confines the analyst is motivated by simulation studies which presuppose experimental data to be sterile, and free of any adaptive processes. However, when adaptive processes are present in real data, the simulation results of Barr et al. are no longer informative. For such data, the method of analysis cannot be purely design-driven, but must be in part driven by the data.
    • "The choice of these cut-off points was a compromise between creating clear, distinguishable motivation profiles and losing the least amount of participants. Despite the wellknown drawbacks incurred in the dichotomization of continuous variables (MacCallum et al. 2002), this practice is herein justified, as our variable of interest (i.e., volunteering status: yes/no) is categorical (DeCoster et al. 2009). In other words, the observation of motives in two groups (volunteers with controlled motivation vs. volunteers with self-determined motivation) allows us to make comparisons with the control group of non-volunteers. "
    [Show abstract] [Hide abstract]
    ABSTRACT: The literature on the health-promoting effects of community work has primarily dealt with the population in retirement age, yet the vast majority of volunteers are people still in the workforce. The aim of this study is to observe the relationship between volunteering and health within the context of working life, considering paid work conditions and motives to volunteer as moderating variables. We conducted an online survey with a sample of Swiss workers employed in different industries. Results show that volunteers with self-determined motives (but not with controlled motives) report lower levels of stress and burnout than nonvolunteers. Moreover, volunteers in general (regardless of the quality of motivation) report higher levels of work engagement and well-being. Analyses further reveal an interaction effect for burnout and stress, where the difference between self-determined volunteers and non-volunteers becomes larger with unfavorable working conditions at their paid job, hinting at potential compensatory effects. Implications for future research and the voluntary sector are discussed.
    Voluntas 09/2015; DOI:10.1007/s11266-015-9647-2 · 0.56 Impact Factor
Show more