Find missing data and understand your results
Check where values are missing, how much of each column is affected, and which records need review. Use this guide to interpret the checker’s rules, percentages and reports before making changes to your data.
Key takeaways
Empty cells count as missing. Enable text markers only when they mean missing in your source.
Percentages depend on included records and selected columns. Display filters do not change totals.
Zero and FALSE remain present by default, even when a value needs a separate validity check.
Patterns can guide investigation. They do not choose a deletion or imputation method.
What is a missing value?
A missing value is a field for which the intended observation is unavailable or has not been recorded. In a survey, an unanswered age question may produce an empty cell. In a sales export, a failed lookup may produce the text NULL. A question that does not apply to a respondent may also be blank, but that blank has a different explanation.
This missing value checker applies the rules you choose. It measures data completeness; it does not decide whether a recorded value is accurate, plausible or suitable for analysis. Start with the source documentation, or use the Data Dictionary Generator to record each field’s meaning and missing-value codes.
How to check missing values in CSV or pasted data
- Upload a UTF-8 CSV or TSV file, or paste a rectangular table from Excel or Google Sheets. Choose the correct delimiter; spreadsheet paste normally uses tabs.
- Check whether the first record contains headings. Preview the import and confirm that values appear under the correct columns. A comma inside a quoted address should remain part of one field.
- Review missing-value rules. Empty cells are always missing here; whitespace-only detection is enabled initially. Text markers require an explicit choice.
- Select the columns relevant to your analysis, then choose Analyze data. Inspect column counts before reviewing individual incomplete records.
- Download the reports and analysis settings so another person can reproduce your choices. Keep an unchanged copy of the source file.
Uneven rows stop analysis by default. Correct the CSV if a delimiter or quotation mark is wrong. If shorter records genuinely omit fields, normalization can add absent cells and retain extra fields as additional columns. It cannot determine where an accidentally omitted separator belonged.
Blank cells, whitespace and text markers
An empty cell has no characters. A whitespace-only cell contains spaces, tabs or line breaks that may look blank on screen. This checker distinguishes those reasons while counting each missing cell once. Trimming is used only for detection; it does not rewrite the table.
CSV has no universal, typed null value. The characters NULL, NaN and None are text until an application interprets them. Here they count as missing only when enabled. An NA code might be a legitimate region label, so the default rules leave it present. Marker matching uses the whole cell: enabling NA does not flag Canada.
Other programs use different conventions. For example, pandas documents that isna() does not itself classify empty strings as missing. Import settings can affect what reaches that function. Document your rules when comparing counts between tools.
Missing percentages by row and column
Column missing % = missing cells in column ÷ included rows × 100
Row missing % = missing selected cells in row ÷ selected columns × 100
Headers never enter these denominators. Complete rows have no missing cells across the selected columns. Excluding an optional notes column can therefore increase the complete-row count without changing any source value. Explorer filters do not change these totals.
Worked missing-value example
Load the verified example above. The labels empty and spaces below describe cell contents; they are not literal input values.
| ID | Age | City | Score |
|---|---|---|---|
| A01 | 25 | Lahore | 80 |
| A02 | empty | Karachi | 0 |
| A03 | 30 | spaces | N/A |
| A04 | 0 | Islamabad | 90 |
| A05 | 28 | NA | empty |
| A06 | 35 | Lahore | FALSE |
Reading the missingness chart and matrix
The horizontal bars show column missing percentages on a fixed 0–100% scale. Use the adjacent table for counts and reasons. The matrix shows records as rows and variables as columns, with P for present and M for missing. It shows the current row window, not a random sample.
Repeated gaps in the same fields may suggest a skipped questionnaire section, a failed import or a collection rule worth checking. The pattern table groups identical configurations and includes complete records. It lists the 15 most common patterns and states how many records the remaining patterns cover. These summaries describe where gaps occur; they do not explain their cause.
What to do before deleting or filling data
First confirm that each marker actually means missing. Then check collection instructions and the original export for import failures or legitimately inapplicable questions. Compare missingness with relevant observed groups, dates or collection methods. The exploratory data analysis guide provides a broader inspection workflow.
Choose any treatment in light of the question, study design and analysis model. Deleting every incomplete record can discard useful information; replacing every blank with zero can change the meaning of a measurement. There is no universal missing-percentage cutoff that makes a column safe or unusable. Document the chosen treatment and consider how results change under reasonable alternatives. Read how to handle missing data before making those changes.
MCAR assumes missingness is independent of observed and unobserved data values under the stated model. MAR assumes that, conditional on relevant observed information, missingness does not depend on the missing values themselves. MNAR allows that dependence to remain. MAR does not mean completely random. A heatmap cannot establish these assumptions, and observed data alone generally cannot distinguish MAR from MNAR without additional information or assumptions. See the National Research Council’s discussion of inference from incomplete data.
Methodology, limits and privacy
The parser handles quoted delimiters, escaped quotes, multiline fields, UTF-8 byte-order marks and trailing empty fields. Values stay as text, including leading-zero IDs and large identifiers. Duplicate or blank headers receive unique column IDs. Data-row numbers refer to records after the header; a quoted multiline record can span several physical file lines.
Blank records stay included unless you choose to exclude records whose fields are all empty. A final newline is a terminator, not another record. An empty physical line within the file is a record; if it has the wrong width, correct it, exclude it explicitly or normalize it. Whitespace-only records are not removed by the empty-record setting.
A separate mask records one reason per missing cell in this order: absent field, empty, whitespace only, configured marker. All selected cells contribute to summaries. Reports use full counts even though the explorer displays 40 records per page and the matrix displays at most 20 selected columns. Percentages with zero denominators are shown as not applicable.
The checker accepts up to 1 MiB of UTF-8 text, 20,000 data records, 100 columns and 100,000 imported cells. It rejects larger input rather than truncating it. JSON and Excel workbook upload are not implemented. Exports quote CSV fields and prefix formula-like values with an apostrophe; normalized absent fields export as empty strings. Exported files are therefore not byte-for-byte copies of the source.
This tool’s code processes data in browser memory, without sending datasets to a server or saving them in browser storage. The shared website may request normal page assets. Close the page or use Clear data to remove the active dataset from the checker. Review exported files before sharing them because the affected-row report contains original values.