I am planning to create groups (low, moderate, high) based on the scores of a psychological scale (i.e. Berkeley Emotional Expressivity Scale-positive expressivity subscale). There is no previously defined cutoff scores for this scale. And it is important to note that I will only use these groups for hypohesis testing purposes (i.e. not for clinical classification as such). I really would appreciate comments and reference to peer-reviewed materials on this issue.
Hi,
In clinical settings, cut-off is often derived from its sensitivity/specificity to differentiate clinical case status. As you say, this will probably not be meaningful in your case. You could define a cut-off so that 1/3 of participants fall into each of the three categories though if that makes sense. In any case, be aware that breaking a continuous variable into groups is often undesirable due to several reasons; e.g. see Dawson & Weiss 2012: "Dichotomizing Continuous Variables in Statistical Analysis".
Best
Dear Cinar,
I defined the cut off score based on the mean of my population:
Article Depression, anxiety, hostility and hysterectomy
A strategy clinical cut offs are often based on is population distribution. For example, moderate could be within one standard deviation of the mean, whereas high is more than one sd above and low is one sd below.
Good question Cinar. Unfortunately, cutoff scores are only as good as the test and the outcome measures used to 'validate' them. If your outcome is not truly dichotomous (present vs. absent), such as any psychopathology, then sensitivity/specificity rates are not valid. For constructs that exist along a spectrum, such as depressive symptoms, very advanced statistical methods and multiple samples are required to form valid cutoff scores. Those statistical procedures ares still under debate and development. So, you can probably get away with any reasonable grouping/cutoffs, but keep in mind that many cases defined as 'high' or 'low' could be miss-classified. In other words, if you give up a continuous measure, such as degree of depressive symptoms, for a rather arbitrary grouping (e.g., 'low' vs 'high') you are giving up a great deal of useful information for a convenient result that will include increased error.
In Grapho- psychology variation to the tune of 5 %20% are present in normal handwriting. Once 2 or 3 trait variations are present simultaneously in same word and lines then it become case of mental challanges of the writer.
in some literature, researchers used quartiles to determine the cutoff points . for example, the highest quartile will represent the high level, the 2nd and third quartiles together means the middle level, and the lowest quartile the low level . this grouping definitely results in some bias . lately, i noticed that, grouping of samples can be based on latent class analysis, which could divide the samples into heterogenous high- middle, or low level group . can such results of latent class analysis serve as a method to determine cutoff points? i do hope that some one can answer the question