BSU FAQ's

1. Do I need statistical help?

Although many researchers in the health sciences have a strong background in statistical methods, it may be beneficial to seek the assistance of a statistician. Many journals require that articles be reviewed by a statistician to ensure that the methods are of high quality for strengthening the design and analytic aspects of the study. In addition a statistician may be able to help with the statistical interpretation and reporting.

2. How many subjects do I need?

This is the question that we are asked most often. The number of subjects needed for a study depends on several factors, including:

· Purpose of the study - Estimation or Significance-testing
· Design of the study
· Variability of the outcome measure
· Significance level (Type I error rate)
· Effect size to be detected
· Power (1-Type II error rate)

3. How do I set up my data for analysis?

In order for us to perform a statistical analysis, we need to be able to read your data. Although many spreadsheets and database management software packages do a wonderful job of displaying and summarizing the data, they are not always easily read by statistical packages. Packages used for statistical analysis are typically set up to read data so that each row forms a case and each column a field (or variable). Thus a data set with 200 subjects and 25 fields will produce 200 rows and 25 columns of data. Typically, column headings (variable or field names) should be limited to eight columns and should consist of letters and numbers. In order to facilitate the statistical analysis, we suggest that you consult with us prior to entering your data into a computer program.

4. What statistical tests should I use for my data?

The appropriate statistical test depends on the study design, on the type of data, and on the assumptions that can be made about the data.

5. Why won't a statistician give me a direct answer?

Statistics involves probabilities and so a statistician is usually hesitant to give a definitive answer. For example, if a statistical test comparing two group gives a p-value of .002, he or she is more likely to say that there is strong evidence of a difference. A p-value gives the probability of the difference observed being due to chance given that a stated hypothesis is true. On the other hand, a non-significant p-value only indicates that there is no evidence of a true difference.

6. Can't you just push a button to obtain statistical results?

Although powerful tools for performing statistical analysis are now available, the statistician needs to consider the design of the study, the nature of the variables, and the assumptions involved in the statistical techniques being used. There are many examples of "horror stories" that have resulted when the underlying assumptions behind statistical tests have been ignored.