rempsyc

rempsyc, 2 months ago

@RPanczak @rstats you can also use this link for the formatted PDF version: https://rdcu.be/dCsjS

rempsyc, 3 months ago

@rstats In particular, nice_table() now converts: (1) objects (e.g., emmGrid) to dataframes; (2) some column names (e.g., “t.ratio” to “t”, sigma to its unicode counterpart).

rempsyc, 3 months ago

@rstats Still for nice_table, (4) z” and “Z” are now formatted to two digits instead of zero digits to comply with the Wald Z statistics in SEM (with lavaan and lavaanExtra); (5) now supports conversion of “CI_lower_B” and “CI_upper_B” columns to “95% CI (b*)”.

rempsyc, 3 months ago

@rstats APA 7th edition actually advises against using beta to represent standardized coefficients, but suggests instead to use a lowercase italic b followed by an asterisk: b*. rempsyc now follows this APA recommendation.

rempsyc, 4 months ago

@lakens @rstats I have a demo here: https://github.com/mdelacre/Routliers/issues/8 where I show that outliers_mcd() behaves well for bivariate cases (1% outliers flagged), but quickly increases its detection rate for 3 variables (14% outliers) or for more realistic numbers in real datasets, like 25 variables (47% outliers). Is this known / expected?

rempsyc, 4 months ago

@AlejandroP @lakens @rstats This is the paper: https://doi.org/10.5334/irsp.289 And the MCD method is strongly recommended over Mahalanobis, but the Mahalanobis distance does not show this "curse of dimensionality" issue in large datasets. If MCD will always flag so many outliers in real datasets, it does not seem realistic to recommend common usage over Mahalanobis? No?

rempsyc, 4 months ago

@AlejandroP @lakens @rstats With the Mahalanobis distance, the outlier detection rates are 0% to 1%, regardless of the number of variables (2, 3, or 25), using the default (recommended) Routliers::outliers_mahalanobis values.

rempsyc, 11 months ago

@rstats - Now produces clearer error messages when specifying columns that don’t exist, across all functions.

• 'nice_table' and 'format_p': Now support the logical 'stars' argument to add significance stars (defaults to TRUE for 'nice_table' but FALSE for 'format_p').

rempsyc, 11 months ago

@rstats 'nice_violin': New arguments; (1) 'order.groups', to order the group factor levels based on value of the "response" variable, or on the string lengths for the group names (factor levels), and (2) 'xlabels.angle', to tilt the labels of the x-axis (useful for long string names).

rempsyc, 11 months ago

@rstats 'nice_violin': For argument 'obs', there are two new choices: "dotplot" (same as 'obs = TRUE' for backward compatibility) and "jitter", useful when there are a lot of observations.

rempsyc, 11 months ago

@rstats 'nice_scatter':

• New argument 'method', with choices "lm" (default) or "loess".
• 'groups.order' now also allows automatic grouping based on value for the response variable (one of the desired levels, “increasing”, “decreasing”, or “string.length”), like 'nice_violin'.