Clean and merge survey data: the errors you can't see in the spreadsheet
An answer cut off halfway. A 99 that means “no answer” and goes into the average as if it were a response. A column that looks numeric and has two values written in words. A case entered twice. None of that shows when you open the file, and all of it changes the results.
You upload the file and within seconds you have the list of what was found. And if you upload two, HILO tells you how they differ and whether it makes sense to merge them.
- The check appears on its own, before you do anything
- Reads correctly the files that Excel opens in pieces
- Never leaves your browser
Four things that silently ruin an analysis
They all show up later, when the number is already in the report.
The file isn't broken: it's being read wrong
SPSS exports wrap values in single quotes, and Excel only understands double quotes. Every comma written inside an answer splits the row, so the file looks like it has extra columns and rows of different widths. Here they are read correctly, and you are told what the difference was.
The 99 that sneaks into the average
99, 999, and -1 usually mean “no answer”. If nobody declares them as missing, they go into the calculation as one more response. They are flagged only when they sit far outside the rest of that column, so a scale that legitimately goes up to 99 isn't flagged. And separately, which columns more than half left blank.
One “don't know” turns the column into text
A single value written in words in a column of numbers is enough for the program to read the whole column as text and calculate nothing. The same goes for decimals: if 3,5 and 3.5 coexist, some are read as numbers and others aren't, depending on the program's language.
The case that was entered twice
Identical rows, and also the column that looks like an ID and yet repeats values. That is how a response entered twice shows up, and you don't see it by looking: you see it by counting.
Before merging, find out whether they can be merged
With two files uploaded, they are compared: whether they have the same shape, in which rows and columns they differ, and whether one has text cut off by an export limit. After that they can be merged, matching columns by the shape of their data and not by their names, because names lie often and data doesn't. Each pair is shown with its reason so you can confirm or discard it.
What it does and what it doesn't
Better to say it here than in the first minute of use.
Statistical analysis
It doesn't calculate, doesn't cross variables, and doesn't make charts. It is the step before: getting the data into shape so the program you use doesn't hand you a wrong number as if it were correct.
Whether two datasets measure the same thing
Before downloading a merge you have to check four boxes: same unit of analysis, same period, same definition, same codes, which only someone who knows where the data came from can do.
The file never leaves your computer
It is read in your browser and uploaded nowhere. In a survey dataset, with names, emails, and informed consent behind it, that matters more than in any other document. And if it also has an open-ended question, you can code it right there without moving the file anywhere.
If the dataset will be shared, you can first remove its identifiers: ID numbers become codes and names are removed, also in your browser.
How the file was read, rows of different widths, whether the header row is missing, fully repeated rows, empty columns, columns where everyone answered the same thing, columns that more than half left blank, missing-value codes treated as numbers, numeric columns with a value written in words, decimals with commas and periods mixed together, and IDs with repeated values.
SPSS exports wrap values in single quotes, and Excel only understands double quotes. Every comma written inside an answer splits the row, so the file looks like it has extra columns and broken rows. They aren't broken: they are being read wrong, and here they are read right.
A 99, a 999, or a -1 that actually mean “no answer”. If nobody declares them as missing, they go into the average as if they were one more response and ruin it without anything warning you. They are flagged only when they sit far outside the rest of the answers in that column, so a scale that legitimately goes up to 99 isn't flagged.
It matches columns by the shape of their data: how much variety they have, what typical length, and how much the vocabulary they use overlaps. If there are headers, an identical name wins. Then it shows each pair with its reason so you can confirm or discard: nothing is merged without you seeing it.
It can paste them together, but the result depends on something no tool can verify: that both count the same unit. One dataset may count people and the other records, and the columns paste together just the same. That is why, before downloading, you have to check four boxes that only someone who knows where the data came from can check.
Excel.
No. The file is read in your browser and never leaves your computer. In a survey dataset with personal data, that matters more than in any other document.
Try it with your own dataset
Upload the file and see what it found. It takes seconds, doesn't ask you to sign up, and uploads nothing.