Assessing a Statistical Analysis

Chi-Squared Test Where Continuous Data Has Been Transformed Into Categorical Data

Using Continuous data as categorical

Statistical analysis can provide a robust approach to assess data, using the raw data to generate useful results which may provide insights into the phenomena being examined. For the results to be meaningful and useful it is essential that the most appropriate method tests are utilized (Bryman & Bell, 2011). In the research undertaken by Razmjou et al., (2009) a decision way made to transform one type of variable, changing it from a continuous variable into a categorical variable. BMI, which is a continuous variable, was changed into categories of normal, overweight, and obese (Razmjou et al., 2009). Usually, transforming continuous data into categorical data results in a loss of detail, so is usually ill-advised, but in some circumstances may help to more useful data (Bryman & Bell, 2011). The circumstances under which continuous variables may benefit from being transformed into categorical variables include conditions under which there may be nonlinearity, for example the presence of a quadratic or exponential curve upwards or downwards, or when you may be a floor threshold effect, or ceiling effect (Bruce, 2009).

When examining the data from the sample, the data from the BMI is likely to reflect a normal distribution of data, where there are individuals in different ends of the scale, those with a low BMI and very high BMI being less frequent than those in the center, but it tends to have a generally linear function, although it has been noted that the distribution in terms of body fat to body mass index is not linear, but has a curve (Bruce, 2009). Therefore nonlinearity is a strong enough reason for the potential loss of data resulting from the classification process. However, while BMI may be continuous, but it does have upper and lower limits of thresholds, and therefore may benefit from categorisation as the results may not be monotonic (Bruce, 2009).

Furthermore, it maybe argued that the division of the sample into the three categories reflects the standard categories which are utilised within the medical profession, with the various categories also having other associated characteristics. Effectively, by dividing the data into these common use categories, the results may be classified in a manner which would be most useful to medical professionals who are using it, with the ability to categorise individuals based on predetermined groups, rather than looking up measures on a linear scale.

Effect on the Data

Whenever continuous data is transformed into categorical data there will be some detail lost. This can impact on the data in a number of ways, firstly, as it is an effective type of rounding, the potential level of probability uncertainty associated with the results may be reduced. However, it also creates sets of data where statistical measures such as Chi square test may be applied, facilitating easier compatibility between different groups, and the differences between groups based on a specific characteristic that the underlying data indicates. Therefore, while losing a degree of detail and accuracy, there is an increased level of ease-of-use and application.

Was it Appropriate?