FIVE QUESTIONS FOR SELECTING A STATISTICAL METHOD

1. How many variables do you have --- just one, more than one, or
way too may?
How many JUST ONCE 2. What is your statistical objective – describe, identify/classify,
variables do compare/test, predict, or explain?
you have? 3. What scales of measurement do the variables use?
4. Are there dependent and independent variables?
DESCRIBE
WAY TOO MANY
5. Are the samples autocorrelated by location (1D, 2D, or 3D) or time/
MORE THAN ONE
sequence (1D)?
Descriptive Statistics
Discrete Scales
Counts, proportions
Conduct a cluster analysis to select Continuous Scales
representative variables Means, standard
Are some CLASSIFY COMPARE PREDICT EXPLAIN
or variable dependent deviations, medians,
Conduct a principal components analysis, and others ranges, quartiles
factor analysis, correspondence analysis, independent? Distribution Fitting
or multidimensional scaling to reduce the No Filtering One-Population [none] [none]
Statistics:
number of variables needed to represent Sorting Test
Yes Skewness, kurtosis
most of the variability Cluster analysis Discrete Scale
coefficient of variation
Graphics: Nonparametric
Histograms, box plots, statistics
Are the Continuous
dependent variables probability plots
Tests: Scales
autocorrelated?
Kolmogorov-Smirnov, t-tests, control
Anderson-Darling, charts
No Shapiro-Wilk Lillifors

DESCRIBE CLASSIFY COMPARE PREDICT EXPLAIN DESCRIBE CLASSIFY COMPARE PREDICT EXPLAIN

Yes
Same as for Cluster analysis, Same as for Discrete scale Discrete scale Same as for [none] Cluster analysis,
Same as for Same as for
single variables discriminant single variables dependent dependent single variables multi
Single variables Single variables
Cross- analysis Multi- variable Logistic variable Logistic Correlations dimensional
tabulations Population Tests regression, regression Discrete Scales scaling, principal
Correlations ANOVA, classification classification Spearman R. components
Discrete Scales ANCOVA, trees, trees, Kendall analysis,
Spearman R. nonparametric discriminant discriminant Tau, gamma correspondence
Kendall tests analysis analysis Continuous analysis
Tau, gamma Continuous Continuous Scales
Continuous scale dependent scale dependent Pearson Product
Scales variables variables Moment
Pearson Product Regression Regression, Correlation
Moment canonical
Correlation correlation

LOCATION DEPENDENT TIME DEPENDENT

Time or
location dependent?

DESCRIBE CLASSIFY COMPARE PREDICT EXPLAIN DESCRIBE CLASSIFY COMPARE PREDICT EXPLAIN

Same as for non- Same as for non- Same as for non- Smoothing
Smoothing Same as for non- Same as for non- Same as for non-
autocorrelated autocorrelated autocorrelated interpolation,
interpolation, Trend-surfaces autocorrelated autocorrelated autocorrelated Time-series
variables with a variables with a variables with a time-series
trend-surfaces, geostatistics variables with a variables with a variables with a regression,
location- location- location- regression,
geostatistics (variogramming time-dependent time-dependent time-dependent ARIMA, spectral
dependent dependent dependent ARIMA, spectral
(variogramming and kriging) variable as a variable as a variable as a analysis
variable as a variable as a variable as a analysis, neural
and kring) grouping factor grouping factor grouping factor
grouping factor grouping factor grouping factor networks