Research Variables and Choosing the Appropriate Statistical Test

Introduction

Understanding the difference between unique types of research variables is critical to evaluating a study's validity and applicability to a particular topic of interest. Similarly, knowledge of different statistical tests and how to select the most efficacious one is an important step in answering a particular research question. The purpose of this paper is to discuss different types of variables found in mainstream research, as well as how these variables influence statistical test selection. Research variables will first be defined, followed by factors that impact different statistical tests. Examples from nursing research will also be provided. This paper concludes with a brief summary and outline of key points.

Research Variables

In any quantitative study, there are generally two primary variables that pertain to the research question. First, there is the dependent variable, which refers to the outcome being measured (Horton, 2009). Most studies begin with a specific dependent variable of interest and base the design around how this outcome is impacted by a range of factors (Horton, 2009). Second, there is the independent variable, which refers to any factor that is manipulated by the researcher to generate change in the dependent variable (Horton, 2009). In a given study, there must be at least one independent and dependent variable, although researchers can include multiple forms of these variables as well. Additional variables that may impact a study are extraneous variables (i.e., environmental factors, separate from the independent variable, which can impact the dependent variable) and moderator variables (i.e., factors, separate from the independent variable, which change the relationship between the independent and dependent variables).

As an example, Testad, Ballard, Bronnick and Aarsland (2010) recently conducted a study seeking to determine the impact of staff training on agitation and the use of restraints in nursing care facilities and home health placements catered toward patients with dementia. In this study, the dependent variables (i.e., the main outcomes of interest) were agitation and the frequency in which restraints were required. To generate change in these outcomes, Testad and colleagues (2010) conducted a two-day educational seminar to regarding effective interactions with distressed dementia patients. Therefore, this seminar served as the independent variables. As described in the study, extraneous variables might have included the location in which the seminar was conducted, the time of day, and patients' psychological and emotional state during the research process (Testad et al., 2010). Furthermore, Testad and colleagues (2010) suggested that the effectiveness of the intervention was largely a result of nursing professionals' knowledge of interaction techniques. Therefore, this knowledge served as a potential moderator variable.

Choosing the Appropriate Statistical Test

Choosing the appropriate statistical test depends entirely on the nature of the variables described above and the number of independent and dependent variables (Horton, 2009). The first factor to consider is the number of dependent variables being measured. Multiple different statistical tests can be performed when just measuring one outcome, although less are available when two or more dependent variables are being assessed (Horton, 2009). In the study described above, the two dependent variables involved necessitated analyzing the variance in mean scores in participants prior to, and following, the intervention. This statistical test is referred to as a "repeated measures analysis of variance (ANOVA)” and allowed Testad and colleagues (2010) to determine the impact of their intervention on multiple participants and several research methods and outcomes.

The second factor to consider when selecting the correct statistical test is the number of independent variables involved (Horton, 2009). In most cases, such as the study described above, just one independent variable is used (Horton, 2009). However, when two independent variables are used, different methods of measurement apply. A factorial ANOVA allows researchers to determine the relative impact of multiple independent variables in isolation, as well as their interaction effect (Horton, 2009). Conversely, a logistic regression test enables researchers to predict the outcome of a dependent variable based on the progressive contribution of multiple independent variables (Horton, 2009).

One final factor to consider when selecting the correct statistical test is the type of data being measured. Data can be: nominal (i.e., in name only, without any sense of order); ordinal (i.e., contains an order, but no definite distance between each data-point); interval (i.e., measurable distances exist between data-points, but not consistent ratios); and ratio (i.e., all the previous data elements exist, as do ratios between data-points). The majority of quantitative research, including the study described above, utilizes interval or ratio data (Horton, 2009). Understanding these types of data is critical because applying a particular test to some levels of data can yield inaccurate results (Horton, 2009). For example, a common mistake made in the social sciences is the application of interval or ratio tests (e.g., ANOVA) to measure ordinal data (e.g., Likert scales). This common mistake can lead to a Type II error, or a false rejection of the null hypothesis (Horton, 2009). In other words, this mistake may lead researchers to believe a particular treatment works, when, in fact, it is not responsible for the change observed in the study.

Conclusion

The purpose of this paper was to discuss different types of research variables, as well as the factors that influence selection of the correct statistical test. There are two primary research variables in any quantitative study, including the independent and dependent variables. Extraneous and moderator variables may also influence a study outcome. Additionally, selecting the correct statistical test depends on three main factors, including: the number of dependent variables, number of independent variables, and the type of data. Understanding these study design issues is critical to preventing several common research errors.

References

Horton, L. (2009). Calculating and Reporting Healthcare Statistics (3rd ed.). Chicago, IL: AHIMA.

Testad, I., Ballard, C., Bronnick, K., & Aarsland, D. (2010). The effect of staff training on agitationand use of restraint in nursing home residents with dementia: a single-blind, randomized controlled trial. The Journal of Clinical Psychiatry, 71(1), 80-86.