# Appendix E- The short R glossary

- Page ID
- 3630

## The short R glossary

This very short glossary will help to find the corresponding R command for the most widespread statistical terms. This is similar to the “reverse index” which might be useful when you know what to do but do not know which R command to use.

Akaike’s Information Criterion, AIC – AIC() – criterion of the model optimality; the best model usually corresponds with minimal AIC.

analysis of variance, ANOVA – aov() – the family of parametric tests, used to compare multiple samples.

analysis of covariance, ANCOVA – lm(response ~ influence*factor) – just another variant of linear models, compares several regression lines.

“apply family” – aggregate(), apply(), lapply(), sapply(), tapply() and others — R functions which help to avoid *loops*, repeats of the same sequence of commands. Differences between most frequently used functions from this family (applied on data frame) are shown on Figure \(\PageIndex{1}\).

arithmetic mean, mean, average – mean() – sum of all sample values divides to their number.

bar plot – barplot() – the diagram to represent several numeric values (e.g., counts).

Bartlett test – bartlett.test() – checks the null if variances of samples are equal (ANOVA assumption).

bootstrap – sample() and many others – technique of sample sub-sampling to estimate population statistics.

boxplot – boxplot() – the diagram to represent main features of one or several samples.

Chi-squared test – chisq.test() – helps to check if there is a association between rows and columns in the contingency table.

cluster analisys, hierarchical – hclust() – visualization of objects’ dissimilarities as dendrogram (tree).

confidence interval – the range where some population value (mean, median *etc*.) might be located with given probability.

correlation analysis – cor.test() – group of methods which allow to describe the determination between several samples.

correlation matrix – cor() – returns correlation coefficients for all pairs of samples.

data types – there is a list (with synonyms):

- measurement:
- continuous;
- meristic, discrete, discontinuous;

- ranked, ordinal;
- categorical, nominal.

distance matrix – dist(), daisy(), vegdist() – calculates distance (dissimilarity) between objects.

distribution – the “layout”, the “shape” of data; *theoretical distribution* shows how data should look whereas *sample distribution* shows how data looks in reality.

F-test – var.test() – parametric test used to compare variations in two samples.

Fisher’s exact test – fisher.test() – similar to chi-squared but calculates (not estimates) p-value; recommended for small data.

generalized linear models – glm() – extension of linear models allowing (for example) the binary response; the latter is the logistic regression.

histogram – hist() – diagram to show frequencies of different values in the sample.

interquartile range – IQR() – the distance between second and fourth quartile, the robust method to show variability.

Kolmogorov-Smirnov test – ks.test() – used to compare two distributions, including comparison between sample distribution and normal distribution.

Kruskal-Wallis test – kruskal.test() – used to compare multiple samples, this is nonparametric replacement of ANOVA.

linear discriminant analysis – lda() – multivariate method, allows to create classification based on the training sample.

linear regression – lm() – researches linear relationship (linear regression) between objects.

long form – stack(); unstack() – the variant of data representation where group (feature) IDs and data are both vertical, in columns:

SEX SIZE

M 1

M 1

F 2

F 1

LOESS – loess.smooth() – Locally wEighted Scatterplot Smoothing.

McNemar’s test – mcnemar.test() – similar to chi-squared but allows to check association in case of paired observations.

Mann-Whitney test – wilcox.test() – see the Wilcoxon test.

median – median() – the value splitting sample in two halves.

model formulas – formula() – the way to describe the statistical model briefly:

- response ~ influence: analysis of the regression;
- response ~ influence1 + influence2: analysis of multiple regression, additive model;
- response ~ factor: one-factor ANOVA;
- response ~ factor1 + factor2: multi-factor ANOVA;
- response ~ influence * factor: analysis of covariation, model with interactions, expands into “response ~ influence + influence : factor”.
Operators used in formulas:

- all predictors (influences and factors) from the previous model (used together with update());
- adds factor or influence;
- removes factor or influence;
- interaction;
- all logical combinations of factors and influences;
- inclusion, “factor1 / factor2” means that factor2 is embedded in factor1 (like street is “embedded” in district, and district in city);
- condition, “factor1 | factor2” means “split factor1 by the levels of factor2”;
- intercept, so response ~ influence - 1 means linear model without intercept;
- returns arithmetical values for everything in parentheses. It is also used in data.frame() command to skip conversion into factor for character columns.

multidimensional scaling, MDS – cmdscale() – builds something like a map from the distance matrix.

multiple comparisons – p.adjust() – see XKCD comic for the best explanation (Figure \(\PageIndex{2}\)).

nonparametric – not related with a specific theoretical distribution, useful for the analysis of arbitrary data.

normal distribution plot – plot(density(rnorm(1000000))) – “bell”, “hat” (Figure \(\PageIndex{3}\)).

normal distribution – rnorm() – the most important theoretical distribution, the basement of parametric methods; appears, for example if one will shot into the target for a long time and then measure all distances to the center (Figure \(\PageIndex{4}\)):

Code \(\PageIndex{1}\) (R):

one-way test – oneway.test() – similar to simple ANOVA but omits the homogeneity of variances assumption.

pairwise t-test – pairwise.t.test() – parametric *post hoc* test with adjustment for multiple comparisons.

pairwise Wilcoxon test – pairwise.wilcox.test() – nonparametric *post hoc* test with adjustment for multiple comparisons.

parametric – corresponding with the known (in this book: normal, see) distribution, suitable to the analysis of the normally distributed data.

*post hoc* – tests which check all groups pairwise; contrary to the name, it is not necessary to run them after something else.

principal component analysis – princomp(), prcomp() – multivariate method “projected” multivariate cloud onto the plane of principal components.

proportion test – prop.test() – checks if proportions are equal.

p-value – probability to obtain the estimated value if the null hypothesis is true; if p-value is below the threshold then null hypothesis should be rejected (see the “two-dimensional data” chapter for the explanation about statistical hypotheses).

robust – not so sensitive to outliers, many robust methods are also nonparametric.

quantile – quantile() – returns values of quantiles (by default, values which cut off 0, 25, 50, 75 and 100% of the sample).

scatterplot – plot(x, y) – plot showing the correspondence between two variables.

Shapiro-Wilk test – shapiro.test() – test for checking the normality of the sample.

short form – stack(); unstack() – the variant of data representation where group IDs are horizontal (they are columns):

M.SIZE F.SIZE

1 2

1 1

standard deviation – sd() – square root of the variance.

standard error, SE – sd(x)/sqrt(length(x)) – normalized variance.

stem-and-leaf plot – stem() – textual plot showing frequencies of values in the sample, alternative for histogram.

t-test – t.test() – the family of parametric tests which are used to estimate and/or compare mean values from one or two samples.

Tukey HSD – TukeyHSD() – parametric *post hoc* test for multiple comparisons which calculates Tukey Honest Significant Differences (confidence intervals).

Tukey’s line – line() – linear relation fit robustly, with medians of subgroups.

uniform distribution – runif() – distribution where every value has the same probability.

variance – var() – the averaged difference between mean and all other sample values.

Wilcoxon test – wilcox.test() – used to estimate and/or compare medians from one or two samples, this is the nonparametric replacement of the t-test.