DNP 830 Topic 2 DQ 2: four data errors and cleaning methods

The verb is *compare*, not list, and it is doing more work than it appears to. Four types of data error described one after another is a catalogue; comparing them means choosing dimensions and holding all four against each one. The useful dimensions are how the error is detected, whether it is recoverable once found, what it does to an analysis if it is left in, and what the cleaning method itself costs. Set that up as the structure of the post and the comparison becomes visible in a sentence per cell. Set it up as four paragraphs and the marker has to do the comparing, which is the part of the task that was being assessed. This is a small structural choice that decides most of the grade. A short table is the most economical way to do it in a discussion post, and it makes the four-way comparison legible at a glance.

Four types with genuine differences between them are missing values, duplicate records, invalid or out-of-range values, and implausible outliers. Missing data is a completeness problem and is invisible in a summary statistic. Duplicates are a uniqueness problem, and they inflate counts and distort denominators without ever producing an impossible value. Invalid values fail a rule you can state in advance — a date of birth in the future, a sex code that is not in the code set, a blood pressure of 900 — so they are the easiest to detect automatically. Implausible outliers pass every format rule and are wrong anyway, which is why they are the hardest of the four and the ones most often left in. Choosing four that fail four different tests is also what stops the cleaning methods from collapsing into one answer repeated four times.

The instruction is to *identify a method for cleaning each error*, and the methods are genuinely different rather than four names for validation. Missing data is handled by deletion or by imputation, and the choice depends on why the values are missing rather than on how many there are. Duplicates are resolved by record linkage, deterministic where a reliable identifier exists and probabilistic where it does not. Invalid values are caught by range, format and referential integrity rules run against a data dictionary. Outliers are handled by returning to the source record to verify them, and only then by flagging, transforming or excluding. One method per error type, each named specifically, is what the instruction asks for. Naming the mechanism behind each method, rather than the software that implements it, is what makes the mapping defensible in a doctoral post.

What lifts this to doctoral level is saying what each cleaning method costs, because none of them is neutral. Deleting rows with missing values throws away the rest of those rows and biases the sample whenever the missingness is related to the outcome — patients who missed a follow-up visit are not a random subset. Imputation invents plausible values and can shrink variance so that findings look more certain than the data supports. Aggressive de-duplication merges two real patients who share a name and date of birth. Deleting outliers deletes exactly the extreme values a clinical analysis often exists to find. A post that names the trade-off for each method is doing analysis; a post that lists the methods is doing recall. Saying which trade-off you would accept, and in what analysis, is a further step and the one that reads as clinical judgement rather than method knowledge.

*Provide evidence supporting your response* means the taxonomy and the methods both need citing, and the informatics literature has done this work already. There are established dimension frameworks for health data quality — completeness, correctness, concordance, plausibility and currency is the best-known set — and mapping your four error types onto a published framework is stronger than inventing categories, because it lets you say which dimension each error violates. It also gives the comparison an external structure rather than one you assembled. Cite the chapter you were assigned alongside the primary literature, since the chapter establishes what the course expects and the literature establishes that the framework is not just this textbook's. That pairing also satisfies the evidence instruction twice over, since one source establishes the framework and the other establishes that it is used.