CADDIS Volume 4: Data Analysis
Basic Principles & Issues
Interpreting Statistics
- Author: D. Farrar
Interpreting Statistics
A complex ecological dataset is likely to exhibit features that can be attributed to variability of data and do not represent stable properties of a system under study. Basic statistical inference procedures (tests and confidence intervals) can help the analyst identify conclusions that are relatively well-supported and associated with acceptable risks of error, taking into account the quantity and variability of the data.
In causal analysis, significance tests and confidence intervals are primarily used to establish whether biological or environmental conditions at a test site differ from expectations, and to help interpret estimates of stressor-response relationships from larger, regional datasets. Other examples of where such inference procedures might be used include:
-
Evaluating whether the form of a stressor-response relationship is consistent among regions, study years, or species.
-
Determining whether one can account for the biological effects of a land-use variable with measurements of proximate stressors.
A confidence interval provides a range for a mean or other parameter that can be viewed as in reasonable agreement with data, taking into account the quantity and variability of data (Figure 1). Statistical tests can be used to avoid any temptation to over-interpret noisy data, by focusing the analysis on effects that are are unlikely to be attributed to variation in the data.
Confidence intervals and statistical tests may be useful in analyzing data to support a causal analysis. However, scoring of evidence in a causal assessment is based on multiple lines of evidence, taking into account statistical and biological considerations. Scoring of evidence is not based mechanically on results of statistical tests or any other statistical procedures.
More information about statistical testing and confidence intervals is available here.