D599 Data Preparation and Exploration, catalog number DTAN 5218, is a three competency unit graduate course in the WGU School of Technology. The catalog describes it as applying analytical programming to the early life cycle steps, cleaning data for structure, accuracy and quality, and interpreting statistics. This is the least glamorous course in the graduate sequence and the one that decides whether everything after it is trustworthy. Every cleaning decision changes the answer somebody will act on later, which is why the defensible version of this work is not the cleanest dataset but the best documented one.
Cleaning is a series of decisions, and each one is arguable
The instinct is to treat preparation as janitorial: remove the bad rows, fix the types, move on. Graduate marking treats it as analytical, because every one of those actions is a choice with an alternative and a consequence.
Take missing values. Whether they matter depends on why they are missing. If the absence is unrelated to anything, dropping those rows costs you sample size and nothing else. If the absence depends on something you did observe, for instance that a field is blank far more often for one shift or one region, then dropping rows removes that group disproportionately and biases the result. If the absence depends on the missing value itself, for instance that high earners decline to state income, then no treatment fully repairs it and the honest response is to say so and describe the direction of the likely bias. That reasoning, rather than the imputation technique, is what a graduate rubric is looking for.
Outliers carry the same structure. An outlier that is a data entry error, an age of 320, is a correctable fault. An outlier that is a genuine extreme, a single enormous order from a wholesale customer, is information, and deleting it because it inconveniences a model is a decision that has to be argued and recorded. The default of trimming anything beyond a threshold is a rule, not an argument.
Then there are the structural problems that decide whether the data is usable at all: duplicate records that inflate counts, a key that repeats when you believed it was unique, dates parsed into the wrong century, categories that mean the same thing spelled three ways, and units that change halfway through a file. Each is invisible in a summary and fatal in a conclusion.
The second half of the catalog description, interpreting statistics, is where the exploration part earns its name. Descriptive statistics are not a formality. A mean well away from a median tells you the distribution is skewed and that the mean is about to mislead someone. A standard deviation larger than the mean on a non negative variable suggests a heavy tail or a contamination. Reading those signals and saying what they imply is the interpretive skill being scored.
Turning scored aspects into a section plan
Scoring detail lives in your Course of Study rather than the public catalog, so open the rubric and count aspects before you touch the data. Each is scored on its own and a score of 2 in each aspect passes the task. Nothing averages, so a spotless cleaned file will not carry an interpretation aspect that reports statistics without reading them.
Head each section with the rubric's own noun, and where code and narrative are both submitted, say which chunk produced the evidence for each aspect.
The word budget, worked. Assume seven scored aspects and directions asking for roughly 2,000 words alongside the code. Reserve 150 for the dataset description and 100 for the close, leaving 1,750 across seven aspects, or 250 each. Then reweight for where graduate marks sit. The aspect on treatment of missing data needs the mechanism, the options and the decision, so lift it to 400. The aspect on interpreting statistics needs a reading of each measure rather than a table, so give it 350. Aspects whose evidence is largely the artifact, such as type correction, drop to 150 of narrative. The total lands near 1,750.
The single most valuable artifact you can keep in this course is a decision log: one row per cleaning action with the problem found, the option chosen, the rows affected and the reason. It answers several aspects at once and it takes seconds per entry while you work.
A structure that fits a preparation and exploration deliverable
Where the task directions specify a structure, follow it exactly. Where the shape is open, this arrangement keeps each decision visible and each aspect easy to find.
| Section | What belongs in it | How it gets scored |
|---|---|---|
| Dataset and purpose | Source, period, unit of observation, initial row and column counts, and the analysis it feeds | Frames the work; cleaning without a downstream purpose has no criteria |
| Initial profile | Types, ranges, distinct counts, missing counts per column, as found | Scored on thoroughness; the before picture makes every later claim checkable |
| Structural issues | Duplicates, key uniqueness, inconsistent categories, unit changes and parsing faults | Scored on detection as much as repair |
| Missing data | Pattern, suspected mechanism, options considered, treatment chosen and rows affected | The central graduate aspect; a treatment with no mechanism discussion is thin |
| Outliers and errors | How identified, which are faults and which are real, and what was done with each | Scored on the distinction between an error and an extreme |
| Transformations | Derived fields, encoding, scaling or reshaping, each with its reason | Scored on reversibility and on whether the reason is stated |
| Descriptive statistics | Measures of centre, spread and shape, with a reading of what each implies | Scored on interpretation; a table with no commentary is an unmet aspect |
| Post cleaning validation | Counts, checks and evidence that the cleaning did what you intended | Scored where named; the step most often skipped |
| Code and references | Script or notebook as directed, plus APA citations for data and methods | Scored on reproducibility |
Evidence craft when the deliverable is a defensible dataset
The standard in this course is auditability. Somebody who disagrees with a finding six months from now should be able to trace it back to what you did and why, without asking you.
- Record counts at every step. Rows in, rows out, rows affected. Three numbers per action is the entire audit trail.
- Never overwrite the source. Read the raw file, transform in code, write a new file, so the original stays available for anyone who wants to check.
- State the rule, not just the result. Removed 214 rows where the transaction date fell outside the study window is auditable. Cleaned the data is not.
- Report the before and after distribution for any variable you treated, so the effect of your decision is visible rather than assumed.
- Cite methods for imputation, outlier detection or scaling to a textbook or paper, and cite package documentation for function behaviour you rely on.
- Keep sensitive fields aggregated or suppressed, and do not paste identifiable records into the narrative to illustrate a quality problem.
The habit that reads as graduate judgment is naming the decision you were least comfortable with. Every real cleaning job has one borderline call, and saying which it was, what you chose and how the result would differ under the alternative is stronger than presenting the process as if nothing was in doubt.
What separates Competent from a submission sent back
Aspects are scored independently, so returns are usually narrow. In this course the frequent cause is a treatment applied without a stated reason, most often a blanket removal of rows with any missing value.
- Every cleaning action names the problem, the option chosen and the number of records affected.
- Missing data is discussed as a pattern with a suspected mechanism, not only as a count.
- Errors and genuine extremes are told apart in writing, and both are handled deliberately.
- Descriptive statistics are read rather than reported, with skew, spread and shape given meaning.
- A validation step shows the cleaned data behaves as intended, with checks that would fail if it did not.
Performance assessment work at WGU can be revised and resubmitted with no grade penalty, so a return costs days rather than standing. In a six month flat rate term those days come out of the courses still ahead of you, and preparation tasks are the ones where a decision log written during the work turns a potential rework into a five minute edit.
Six mistakes that cost time in D599
- Dropping every row with a missing value. It is fast, it is often wrong, and it is the single most returned decision in preparation work.
- Imputing without saying what it does. Filling gaps with a mean shrinks variance and can hide the pattern the analysis was meant to find, so the trade off has to be written down.
- Treating outliers as noise by default. Some of them are the most informative records in the file.
- Trusting parsed dates. Two digit years, mixed formats and locale differences all produce dates that look plausible and are wrong.
- Skipping validation. Cleaning that is never checked can introduce faults, and the check costs one code block.
- Reporting statistics without reading them. The catalog names interpreting statistics explicitly, so a table with no commentary leaves a named aspect unanswered.
How support works on this course
Send the rubric from your Course of Study, the task directions and the dataset. The work comes back aspect mapped and audit ready: an initial profile taken before anything changes, structural problems found and named, missing data discussed as a pattern with a mechanism and a justified treatment, outliers separated into faults and real extremes, transformations with reasons, statistics interpreted rather than tabulated, and a validation step at the end. The decision log comes with it, which is the artifact that makes the next course in the sequence faster.
Where a proctored objective assessment sits on this course, the boundary is fixed. Proctored exams are yours to sit. We prepare only, with practice datasets, drilled statistical interpretation and an honest go or wait read. We never sit assessments and we never ask for portal credentials.
Questions students ask about D599
Is D599 the same course as DTAN 5218?
How is D599 different from a data wrangling course?
Which language should I use for the cleaning work?
Cleaned the data but cannot defend the choices?
Send your rubric and dataset. You get an aspect mapped draft with a decision log, before and after distributions and statistics that are read rather than listed.
Where D599 sits in WGU's programs
The July 2026 catalog places this code in 3 current WGU programs. Open a program page for the complete standard path and term positions. The live Degree Plan remains authoritative after transfer credit, substitutions, and mentor planning.
The assessments, one by one
The public catalog does not publish this course's PA/OA identity or task count. WGU Tutors publishes at most one PA manual per course and only from a WGU-controlled public rubric. Until that source exists, PA help begins from the student's real Course of Study and OA support remains preparation only.