The ghost of selective inference in spatiotemporal trend analysis

Area of Science:

• Environmental science
• Geostatistics
• Statistical analysis

Background:

• Spatiotemporal trend analysis often involves numerous statistical tests on gridded data.
• Selective inference, focusing only on pre-defined significance thresholds (e.g., p < 0.05), ignores the total number of tests performed.
• This leads to multiplicity problems, inflating the chance of false discoveries.

Purpose of the Study:

• To review the challenges of multiplicity and selective inference in spatiotemporal trend analysis.
• To discuss methods for managing inflated spurious results in gridded data.
• To promote transparency and replicability in scientific studies.

Main Methods:

• Review of existing literature on multiplicity and selective inference.
• Discussion of statistical adjustment methods for p-values.
• Illustrative examples using gridded data in trend analysis.

Main Results:

• Uncorrected multiplicity and selective inference can create illusory significant trends.
• Rigorous correction methods are necessary to reveal true statistical patterns.
• Statistical significance is dependent on the entire set of tests performed, not just those below a threshold.

Conclusions:

• Properly addressing multiplicity is essential for reliable spatiotemporal trend analysis.
• Transparency in methods and reporting enhances study replicability.
• Understanding the full context of statistical tests, including non-significant ones, is vital for accurate interpretation.