rhoinc / sas-codebook

A SAS macro for generating a concise summary of every variable in a SAS dataset.

sas-codebook

This repo contains SAS^® code for producing a PDF summary of each variable in a dataset. Each variable summary is confined to a 1" tall strip, allowing for 9 variables to be summarized on a page.

Metadata is provided for each variable in bold text. This metadata includes the variable's name, label, type, and (if applicable) format.

Below the metadata are the number of non-missing values, the number of missing values, and the number of unique values.

Continuous variables are summarized graphically using a combination histogram/boxplot. Continuous variables are also summarized tabularly using the mean, standard deviation, min, max, median, and quartiles.

Categorical variables are summarized graphically using horizontal bar charts (or dot plots). Categorical variables are also summarized tabularly using counts and percents.

For more details about the SAS code see the Wiki.

sas-codebook's Issues

evaluate potential for a 4th macro: two-way comparisons

Primary motivator is a treatment/control type comparison.

more metadata [3]

Consider adding more high-level information at the top of each report related to the dataset. For instance:

• Dataset label or descriptor
• Replicate some R package that summarizes datasets at a high level (pct of missing obs, for instance)

panels: possibly change histograms to variable-width box plots [6]

panels: add "obs by panelby" at top [7]

Add a table at the top of page 1 which displays the number of observations in the dataset for each level of the panelby variable. Will likely want to match the sort of the subsequent plots, but this is not a certainty. Might display counts as text or bar charts.

generic: add appendix of plots [?]

Not sure if there's much added value here, but maybe a hyper-visual individual would find value in it.

prevent PNG files from cluttering directories

Spencer just got 3700 PNG files - ick!

panels: deal with missing values in PANELBY variable

fix letput issue

add %nrbquote to deal with apostrophes

vertical: long VAR= list causes SAS error

When the VAR= list is long enough to cause multiple pages of output within each BY= level, SAS crashes. This is apparently an issue because of the use of STARTPAGE=NOW at the top of each BY= level. Because I am using STARTPAGE=NOW, SAS gives up trying to control page breaks. If there are a lot of variables (more than will fit on a page), it crashes. 😢

Short-term hack for users: don't specify more than 8 variable names in the VAR= parameter. Call the macro multiple of there are more than 8 variables of interest.

Possible long-term workarounds include:

• Control all page breaks by making assumptions about the heights of headers and strips.
• Possibly rewrite the right half of the report to be an annotated image (thereby controlling the height).
• Possibly rewrite the header to be an annotated image, again to control the height.
• Both of the above would likely depend upon the experimental OUTFILE= option.

vertical: add summary of by levels at top of report [5]

Complete list of all combinations of the by variables and count of obs within each level. Could run into width issues if lots of variables are specified. Might need to go landscape for this first page and then flip back to portrait for subsequent pages.

panels: possibly work around latency with one-page-at-a-time rewrite [2-8]

Latencies never seem to appear on page 1, so perhaps create several one-page outputs and stack them together on the back end. Would lose the automatic page numbering, though could possibly manually number the one-at-a-time pages as they are created. Getting the "of Y" part would require more pre-planning/measuring/counting, but again still possible.

generic: add appendix of all (unformatted = formatted) values

To appease those who would say "that's not a codebook", consider adding an appendix at the end which shows more than just the "Top 5" that are shown in the 1" strips. Would like include counts and maybe percents. Might still need to include some sort of max for cases with 100+ unique values.

If this is implemented, consider including hyperlinks so that folks could jump back and forth from the strips to the appendix and back. Not sure how to do this, but how hard could it be!?

panels: distinguish sample size somehow [8]

Distinguish sample size at each level of the panelby variable.

• Sort by number of obs
• Shade by number of obs

Graphics Don't Output to PDF

I'm running the macro "codebook_generic" and the PDF looks great, but the graphic for each variable doesn't end up on the PDF. Only the graphic for the first variable appears. The graphics just get output to PNG files in my data folder. I've only specified the macro variables "DATA" and "PDFPREFIX". I'm using SAS/STAT 14.1. Do you know what might be going on?

wiki: transform TracingThrough.docx into something resembling a lifecycle writeup [5]

generic: deal with long variable names

The macro creates secondary versions of variables by adding prefixes (e.g., VAR1 becomes CB_CHAR_VAR1). This is problematic when incoming variable names are already approaching the length limit of 32. Need to update the code to somehow get around this issue.

Log messages create lag between codebook outputs and program log.

Running the codebook on multiple datasets, interactively on SAS Grid through EG at least, creates a substantial lag of nearly a minute per dataset. Recommend substantially paring down messages sent to the log.

Allow Vertical implementation of Paneled codebook

Low priority query from DMEDS

asap: verify whether or not the paneled version is still buggy in 9.4m3