Why should you avoid using point-and-click method in statistical software packages?
Many statistical software packages provide one of two alternative methods for users: a method based on using command language, or a "point and click" method. Users may choose either one of these methods to manipulate/analyze their data, and, in most cases arrive at the same results. Some users find the command language option very complicated to follow and chose to go with the point and click method. There are, however, some important ethical and practical issues one should consider before using a point-and-click method in statistical computation.
Replicating study findings is a necessary and desirable aspect of science. In empirical research, there is a reasonable expectation that the results of empirical work or of experiments can be replicated by another researcher using the same data.
Some professional journals, for example, will not publish empirical work unless the researchers can provide information that would be sufficient for another researcher to replicate the results. Because it is very likely that the raw data are transformed many times during a data analysis period (e.g., dropping some cases because they are "outliers," or manipulating some variables for theoretical reasons) the researcher has to keep a very detailed record of each and every decision made along the way. This is very difficult to achieve using a point-and-click method. For example, in SPSS, if one deletes a case or transforms a variable and then saves the end result, one can not go back and redo the analysis unless they save the original data-set in another file. Hence general standards of research ethics require that researchers make notes on what they did in every step of the analysis, and keep copies of computer programs (e.g., a copy of syntax program in SPSS) and data sets. Solely using point and click method could put the researcher at odds with these research ethics.
Other advantages of using the program's command language (e.g., SPSS' syntax method):
These arguments against the point-and-click method also apply even in the context of a package with a command language facility (e.g., Stata, SAS). Writing a program does not mean writing singular commands. That is, giving commands ad-hoc is only good for exploratory analysis. In order to create a command program, one is expected to develop a data analysis schema, which requires detailed preliminary thinking about how to handle the data.
For more information on this issue:
We would like to thank Mick Smyer, Donald Cox, Jenny Baglivo, Nicholas J. Cox and Petia Petrova for their comments on earlier drafts of this statement.
Christopher F. Baum
Graduate Statistical Assistant Program, FMRC