Errors in Spreadsheets are Pandemic

G Roper writes "Studies show that most spreadsheets have critical errors in one percent of their cells, well beyond a permissible level. Here are some news stories about spreadsheet errors. Spreadsheets won't protect a firm from liability when they are audited and spreadsheet errors found: spreadsheets are not secure, provide no audit trail and won't pass HIPAA or Sarbanes-Oxley auditing. How are Slashdotters coping with the proliferation of spreadsheets in the face of greater legal accountability and auditing?"

Comments

Easy question.

[Anonymous Coward]

"How are Slashdotters coping with the proliferation of spreadsheets in the face of greater legal accountability and auditing?"

With a pencil. haha.

I hear we need:

[j2crux]

Well as one of my bosses says, "We need more Double E masters."
Alas he doesnt mean Electrial Engineers, but "Excel Experts."
He's very bitter about his education :P.

Re:I hear we need:

[BMonger]

I hear they have some great "Excel Expert Masters" in the department of redundancy department.

Re:I hear we need:

[smokeslikeapoet]

No kidding. I had to compute 5 year straight-line depreciation for accounting records going back to 1992, to insure our estimates were correct. I wasn't going to compute 10 different formulas for each year that also omitted equipment sold during those years. The result was a 20 line excel formula that required a little programming prowess. I'm the IT guy here and don't normally do accounting stuff, but it wouldn't have gotten done by our deadline without me. Most accountants do not have the programming knowledge to use spreadsheets effectively. And most schools do not teach spreadsheets at the level that is needed in the real world. It really is a forgotten realm in the industry. Oh yeah, and F1 really helps when you get stuck.

spreadsheet errors are hard to fix

[yagu]

From the abstract: "Although spreadsheet programs are used for small "scratchpad" applications, they are also used to develop many large applications. In recent years, we have learned a good deal about the errors that people make when they develop spreadsheets. In general, errors seem to occur in a few percent of all cells, meaning that for large spreadsheets, the issue is how many errors there are, not whether an error exists. "

I think "how many errors, not whether an error exists" is just as true for applications and programs written in any language or using any technology. What's so insidious about spreadsheets is their integrity and the difficulty to maintain that.

Once you start changing any complex spreadsheet you risk and almost guarantee corrupting other parts of the spreadsheet ostensibly okay. The spreadsheet is so inextricably integrated to itself, you pull one string, and some widget a million miles away suddenly misbehaves, though, you're unlikely to notice until later, if at all.

IT should be strict about policy around spreadsheets... spreadsheets are great powerful tools, but they shouldn't be anointed as applications.

I worked on a team that created a large software development workbench. A critical piece of this workbench included a suite of spreadsheets with amazingly complex macros and formulae hidden way out of the casual users' sight. Immediately upon release (and much aligned with my warning and prediction) the workbench fell apart on a daily, even hourly basis, among many teams out in the field. Turns out users were deleting rows in the template spreadsheets deemed irrelevant and unnecessary to their work. Guess what got deleted along with the "unnecessary rows"? Yep, chunks of macros critical to the proper function of the workbench.

Re:spreadsheet errors are hard to fix

[morgan_greywolf]

Once you start changing any complex spreadsheet you risk and almost guarantee corrupting other parts of the spreadsheet ostensibly okay. The spreadsheet is so inextricably integrated to itself, you pull one string, and some widget a million miles away suddenly misbehaves, though, you're unlikely to notice until later, if at all.

Well, as you alluded to earlier in your post, whether a spreadsheet has errors in it depends on how it was made.

This also goes for maintaining integrity of the spreadsheet. Both OpenOffice.org and Microsoft Excel offer the ability to protect cells from modification. If you design your spreadsheet application in a certain way, you can prevent corruption to the spreadsheet through modification. It's tricky and it often requires a lot of macros and workarounds to make it happen, but it can happen. Also both Excel and OOo offer the ability to track changes made by users, so there is some level of built in accountability -- but not much.

One of the main points of TFA, I think, is that spreadsheets are good for quick-and-dirty scratchpad applications, but really fail to complex applications that require maintainability, documentability, and good authentication and security surrounding changes.

If you need that, you need a database application. This is what I've been telling people for YEARS -- don't use Excel for what you really need a database app for, and, conversely, don't write a database app for what you could easily just as easily do in Excel.

Re:spreadsheet errors are hard to fix

[heinousjay]

Type-o's?

Negative.

Re:spreadsheet errors are hard to fix

[bill_kress]

I think "how many errors, not whether an error exists" is just as true for applications and programs written in any language or using any technology. What's so insidious about spreadsheets is their integrity and the difficulty to maintain that.

The answer to that question is that Spreadsheets are not designed for maintenance like most languages are. The difference here is HUGE, in fact, there is no comparison whatsoever.

Applications designed by real teams in real languages involve some absolute requirements:

Here are some bare minimal language issues that most decent developers wouldn't question for work on a team (Individual developers/web developers are often a bit more loose)

Data hiding... CRITICAL! Any language that defaults to global data would be ludicrous. To even allow data access beyond the smallest boundaries is frightening.

Code organization--you should be able to group common functions with the data that they represent.

Code History--the ability to compare code changes to a previous version.

Highly documented code/self documenting code--duh

Some form of design--many large projects are in/moving to OO code, it's difficult to handle the design of a large application without it.

I'm not saying these things are impossible to see in a spreadsheet, but pretty unlikely. On top of that, the level of freedom given to the user of a spreadsheet makes the data environment of the program extremely difficult to control.

Absolutely opposite ends of a spectrum.

Hardware?

[punkass]

Ok, slightly off topic, but why is this posted in Hardware?

Re:Hardware?

[j0e_average]

Actually, the editors had all the stories queued up for the day in Excel. They managed to sort the list incorrectly, which caused this story to be posted under Hardware by mistake.

Re:Hardware?

[Darkman, Walkin Dude]

I also demand to put the scores near the comment title.

I think the thinking there is to cut back on the karma whoring and make comments stand on their own merits. Also it should help keep groupthink under control, and is more indicative of the fact that moderation really only represents the opinions of one, two or maybe five basically random people out of all the thousands that read slashdot. To whit, its not terribly important.

Re:Hardware?

[_xeno_]

Assuming you're using Firefox 1.5 or higher (or Seamonkey 1.0 or higher?), I've created some CSS rules to make Slashdot use a Serif-style font and move the comment score below the title. I would have just copy-pasted the rules directly into the comment, but Slashdot's stupid broken <ecode> tag bravely mangled all attempts. So instead you'll have to live with a link to the rules on my personal site - no, I'm not spying on you. :)

(Why move the scores below the title and not next to the title? Because my attempt to move it next to the title didn't look quite as nice as I'd like thanks to the current setup. At some point in the future I may create a set of rules to move it to the right of the title, but it's going to be a long set of rules.)

ODF

[From A Far Away Land]

If every change even a correction needs to be audited save-to-save of a file, then why don't we implement a Wiki style log of changes to the file? I wonder if Open Document Format would easily support this.

The mountains of next-to-worthless data the piles of auto-saves would generate is mind boggling.

Re:ODF

[Skreems]

For god's sake, please, PLEASE let them not cram yet another change tracker into a format that shouldn't support it. Change management already exists in so many forms it's not even funny (cvs, svn, source safe, etc), and works off the shelf with any document format. If people would just learn to put existing tools together instead of shoehorning all functionality into every application, things would be a lot simpler.

Probably not very well..

[Ckwop]

How are Slashdotters coping with the proliferation of spreadsheets in the face of greater legal accountability and auditing?"

My guess it they're not. I've met FIERCE resistance in the past from accounts trying to reform their spreadsheet ways. Every accountant understands the spreadsheet. The Financial Director understands the spreadsheet. If you can't get the Financial Directory to back your plans then any reform is dead in the water.

The problem is born out of bad communication skills. IT generally assumes that just because the FD doesn't understand C++ he is stupid. We see this kind of behavior all the time on Slashdot:

"What amazed me is that the Judge really understood the GPL."

No fucking shit he understood the GPL. Let's see he probably got a 1st class degree in Law, Passed his BVC with flying colours. He then probably got his pupillage with ease (there are twice as many students each year as there are pupillages) and then rose to the Bar. After that, he'd have spent 15 years working cases in the Crown Court. If he didn't understand the GPL he would have fallen at the first hurdle. My brother is a lawyer and understood the GPL before he even took his LPC. By comparison, you're average IT guy is a mere peon. I'd wager that given your average programmer with no C++ experience, the Judge could beat the programmer hands down in a programming contest. These people are very, very smart.

The same is true of Financial Directors and their ilk. They have to take years of qualifications and have decades of experience before they're allowed to do their job. Talking down to them is a recipe for marginalization. So the solution is to talk to them in clear language. None of this bullshit bingo that seems to be infesting every cranny of IT - clear, plain language.

Explain the problem, then explain the solution. They don't want or care to hear about LAMP, AJAX or Web 2.0. This like a builder telling you the type of screws he's going to use to build your house. All that you care is that your house is well built and will last a long time without significant maintenance. All they want to know why they need your solution, how much it will cost and the consequences if they don't do it. Anything else is a waste of their time and will lower the amount of time they have for you.

Simon

Re:Probably not very well..

[pla]

None of this bullshit bingo that seems to be infesting every cranny of IT - clear, plain language.

I largely agreed with you right up until that line...

You compain about IT playing "bullshit bingo", compared with judges and financial guys?

IT may overnominalize, but (unlike law and accounting), we tend not to completely redefine perfectly good words for our own uses. Learning what a TCP/IP stack does takes some effort, but once you know the phrase, you know the phrase.

By comparison, every time I get into an argument with a law-geek and they play the "but that word doesn't mean the same legally as it does in English" card, I just want to serve up some serious hurting.



Now, I agree that judges and CFOs most likely understand the apparent BS they speak fluently. But don't try to complain about geek jargon as magically worse.

Re:Probably not very well..

[crankyspice]

By comparison, every time I get into an argument with a law-geek and they play the "but that word doesn't mean the same legally as it does in English" card, I just want to serve up some serious hurting.

The 'problem,' as I see it, is that the law demands exactingly precise use of language. (I've personally witnessed multi-million dollar litigation over the position of a comma, because it changed the meaning of a sentence.) The legal use of language tends to be unerringly precise -- as precise as, say, C demands you to be. Most English speakers use English more fluidly; think "Perl," to continue my programming language analogy.

If you can give me a term (or terms) that you've encountered that has a 'different' legal meaning than it does in common conversational English usage, I could speak more intelligently to this point. I suspect, however, that an analysis of the true definition and etymology (check with Black's Law Dictionary -- 6th Edition if you can find it, though even the 8th has merit -- and the Oxford English Dictionary) will reveal that the legal usage is the proper usage, at least historically. (As to why legal terminology doesn't change to reflect common usage -- I'd guess stare decisis; it's not uncommon to cite to legal opinions or treatises that are a hundred years old or more; the words have to have the same (legal, not conversational) meaning today as they did then, or the whole mess gets way, way too confusing.

Re:Probably not very well..

[DragonWriter]

The 'problem,' as I see it, is that the law demands exactingly precise use of language. (I've personally witnessed multi-million dollar litigation over the position of a comma, because it changed the meaning of a sentence.) The legal use of language tends to be unerringly precise -- as precise as, say, C demands you to be.

The problem -- and I say this as a law student and hobbyist programmer -- is that lawpeople are more able to isolate themselves from C, and that C (unlike, say, COBOL) doesn't borrow as heavily from everyday English, so that more people can avoid dealing with C's quirks than can avoid dealing with "Lawspeak", and if you run into a pile of C code, it won't look like a bunch of English that you should be able to understand. Plus, usage in law is a lot more fuzzy and subjective than that in any programming language, since courts generally are not Turing machines. (Plus, legislators don't tend to have much of a structured design methodology, or even much of a concept of a "system" that is being modified, and there is virtually no analog to "testing", so most law is really bad "code" that implementors -- courts as well as executive agents -- are forced to rationalize on the fly.)

If you can give me a term (or terms) that you've encountered that has a 'different' legal meaning than it does in common conversational English usage, I could speak more intelligently to this point.

There's a lot -- "person" can be a fun one. Of course, lots of terms don't just have a distinct "legal" definition, but they have very specialized and conflicting definitions in the context of particular laws, such that (to mix usages from different fields) PatriotAct::terrorism may not be defined the same way as AntiTerrorismActof1996::terrorism. Of course, this shouldn't be hard for those familiar with programming to understand, though the fact that the resolution of these kind of conflicts is often implicit, and relies on familiarity with a "code base" of law that most people aren't that familiar with, plus there is a lot of misleading information out there which says "In the law, X means Y", when that's only true in a particular context within the law, not in "the law" more generally.

As to why legal terminology doesn't change to reflect common usage -- I'd guess stare decisis; it's not uncommon to cite to legal opinions or treatises that are a hundred years old or more; the words have to have the same (legal, not conversational) meaning today as they did then, or the whole mess gets way, way too confusing.

Plus, like any jargon, its pretty clear -- at least moreso for the concepts involved than "everyday English" would be -- to the people who are deeply involved in it, and for them the context switch is fairly seemless, so there is no real need to change. Meanwhile, well-meaning attempts to force "common use" into legal language end up with results that confuse everyone involved, and usually result in creating more specialized legal jargon terms that are subtly different from the old ones, have to be used along with the old ones, and still don't correspond to common usage.

This should be obvious

[Rude Turnip]

"How are Slashdotters coping with the proliferation of spreadsheets in the face of greater legal accountability and auditing?"

I don't know about you, but I actually check my work and co-workers cross-check each other's work. Any spreadsheet whose numbers can't easily be checked out on a calculator should be designed such that the information generally flows in one direction and each step of a calcuation is broken out into separate rows whenever possible to make "debugging" easier.

Spreadsheets fundamentally flawed

[sfraggle]

Spreadsheets are basically a form of visual programming language, so it is unsurprising that bugs occur. They are basically designed so that ordinary people can use them, which means that they lose some of the strictness that is enforcable in a normal programming language. More worringly, I'd say that some of the properties of spreadsheets naturally encourage bugs. For example, when programming, code duplication is considered bad, and shared common code good, because it encourages simplicity and when bugs are found, they can be fixed in a single location. Conversely, in spreadsheets, the user is actually encouraged to duplicate code, with tools that let you "drag down" equations into neighbouring cells. Perhaps we should be wondering if it would be a good idea to create some kind of "next-generation" spreadsheet system that addresses these problems. Whereas programming languages have evolved constantly over the years, spreadsheets remain unchanged.

Re:Spreadsheets fundamentally flawed

[kebes]

I agree. Spreadsheets are quite powerful, but I'm often frustrated with them. Since I have programming experience, I can see the ways that spreadsheets could be made more robust and more powerful.

The first thing to change is what you alluded to: code should not be duplicated, but linked instead. When you drag a formula, it should really just fill those cells with references to the formula to be used. When you try and edit any one of those cells, you are given a popup where you can edit the master equation used in that range. This would make it so much easier to fix spreadsheets. With fewer points of failure, it is much easier to find bugs or add functionality.

A related point is that the way a single cell is designed makes it hard to read complex equations. A complex operation should generally be split across multiple cells, as this makes debugging and understanding workflow easier. However sometimes you need a single cell to be quite complex, and the way most spreadsheets display the cell contents (as a single long line) makes it difficult to understand. Again the cell contents should appear in a pop-up, where proper indenting, bracket-balancing, comments, and color-coding can occur (i.e.: everything that a normal programming IDE gives you).

Another thing that would make spreadsheets more useful/powerful would be the ability to COMPILE them into another form. I often use spreadsheets for prototyping a new analysis, and then re-code it into another form (Java, C++, Matlab, etc.) for efficiency purposes. In many cases this is a good idea, since it makes sure the programmer understands the problem fully. However in other cases it is wasted effort. A spreadsheet is slow to calculate but sometimes it provides the best layout for coding a solution. What I would like to see is a spreadsheet program that converts the entire spreadsheet into some kind of human-readable linear code (C++ style syntax or whatever). This would involve converting blocks of numbers into vectors, arrays, or matrices, automatically naming them (based on the column header, for instance), and creating loops to account for iterative operations, and translating all the spreadsheet functions into other types of syntax. Having this human-readable version of the code would be great. It could be fixed and improved (for efficiency or interacting with other programs), commented, and so on.

This human-readable code could then (obviously) be compiled into an efficient binary form. This would make spreadsheet concepts of workflow applicable to more demanding applications.

Lastly, I think spreadsheets need to learn what other programming forms already know: comments are important! The spreadsheet should strongly encourage the user to enter an explanation for every formula they write. Everything should be commented. This is the only way for future people to fix or modify the spreadsheet. Plus, accountability and traceability are easier.

Perhaps I'm asking for too much... but I think if spreadsheets evolve in this direction (towards being a more rigorous programming environment), the benefits would be huge. People are now (more or less) used to using a spreadsheet. This kind of "programmer's spreadsheet" would be great for people who know programming (it becomes more powerful) and also for casual users (some rules enforce better practices).

Auditing in Excel

[everphilski]

provide no audit trail

You can provide an audit trail in Excel:

Tools->Share Workbook->click "Multiple Users"->click "Advanced"->select how many days you want to keep a history for.

(It might not be good enough for HIPAA or SA but there is an audit trail ... )

ever heard of locking cells?

[jbeaupre]

Sounds like you're advocating the wrong policy. How about locking the cells so users don't screw things up? You wouldn't let non-programers alter code, why would you let them alter the spreadsheet?

Re:ever heard of locking cells?

[baldass_newbie]

Even then you're still subject to VBA rounding issues which even most programmers are woefully ignorant of.
I mean locking cells is nice. Understanding the functions you're using is even more important.

Re:For one thing, don't use Excel

[winkydink]

What color is the sky on your planet?

Mod Parent Up!

[mrchaotica]

Yet another example of the truth of "those who do not understand UNIX are bound to reinvent it, poorly."

Re:and the error rate before the computer age....

[PlusFiveTroll]

I'm not sure where you got the %1 idea from, but in one of the linked articles there was a a $50 million dollar spreadsheet error (spending bugeted money that did not exist). There was also an error in a spreadsheet that miscalculated natural gas reserves that causes a BILLION DOLLAR rise in the commidity value (aka speculators) which was not real.

and lastly, who cares? Think Sarbanes-Oxley, if your a CEO, you care, alot.

Next up: Bugs in Computer Programs are Pandemic

[Shimmer]

Film at 11.

Minimize the errors

[MobyDisk]

It is difficult to eliminate the errors, so a better solution is to minimize them. The easiest way to do this is to add extra workbooks named "sheet2" and "sheet3" with thousands of extra cells in them. Then, the percentage of error is 3 times lower. Example:

Before: "sheet1" has 50x50 cells, with 25 errors. That's 25 / 50^2 = 1% errors.
After: Add "sheet2" and "sheet3" with another 50x50 cells. Now, the error rate is 25 / 50^2 / 3 = 1/3 % error.

According to my spreadsheet, that is a much better error rate!

Sum of Misspent Years

[Doc Ruby]

I'm amazed at the rut Excel traps IT-based businesses in. Excel was once the best thing computers did, apart from screensavers and ahead of email. But by the late 1990s Excel should have become merely the GUI for relational databases. Even cheap/free ones like MS-Access and MySQL, if not Oracle, Postgres, SQL-Server. Excel should have had macros programmable in the exact same language as actual databases, like VB (not VBA), Perl or something unique to its vertical integration. Upgrading from the starter DB to the enterprise DB should have been a matter of installing the new backend on the network, and configuring the Excel client.

If that path were taken, Excel would be a manageable platform. Instead, it's trapped in the early 1990s desktop, with all its limitations to collaboration, performance, maintenance and dataflow. Every improvement in those areas is a one shot deal, a hack on a once-elegant app now hacked to death.

Maybe the new generation of open formats and distributed computing services offer a chance to try again. Excel will probably include those, just diluted by all the wrong ways retained as its "legacy".

Can't resist

[isotope23]

I will give you what I consider the most blatant and insane example.

U.S. Constitution article III

The trial of all crimes, except in cases of impeachment, shall be by jury; and such trial shall be held in the state where the said crimes shall have been committed; but when not committed within any state, the trial shall be at such place or places as the Congress may by law have directed.

U.S. Constitution: Sixth Amendment
Sixth Amendment - Rights of Accused in Criminal Prosecutions

In all criminal prosecutions, the accused shall enjoy the right to a speedy and public trial, by an impartial jury of the State and district wherein the crime shall have been committed, which district shall have been previously ascertained by law, and to be informed of the nature and cause of the accusation; to be confronted with the witnesses against him; to have compulsory process for obtaining witnesses in his favor, and to have the Assistance of Counsel for his defence.

Now to me and I think to 99.999% of americans the phrase ALL clearly means every single instance.
To judges and lawyers however this is apparently different. As it currently stands you do not have the right to a jury trial.

Sometimes, it's *not* an error...

[rmckeethen]

About 10 years ago, a Silicon Valley manufacturer of medical imaging equipment hired me to do accounting work for them. Among my many tasks for this firm was the weekly generation of a report based on the company's current accounts receivable balance. I was told that this report was very important since it was used by one of our execs. during his weekly 'power breakfast' meetings with the other heads of the company.

A month after I arrived at the company, I noticed that the numbers didn't look right when I generated this weekly report. I started examining the spreadsheet formulas and soon found a small error in one of the calculations we used to derive our total balances. I notified my manager and we both agreed that the original spreadsheet wasn't giving accurate results. I corrected the formula and then patted myself on the back -- after all, I'd uncovered an error that many people, including my manager, had missed for months. I thought I was in good shape at the company after that because I'd done the right thing. I'd fixed a problem. Yay for me.

However, a week later, my manager brought me into his office to talk about the issue. I was more than a little surprised when he asked me to go to my desk and change the formula back to what we'd used before. I asked my manager if he still agreed with me that the old formula was giving incorrect data. He just smiled and said yes, he agreed with my original assessment. I was right, he told me, but our exec. had still asked him to revert to the old formula, no reasons given.

Shortly after this incident, my manager again brought me into his office. He had a pained look on his face as he began to tell that the company wouldn't be needing my services anymore. My manager never gave me an explanation as to why, but I didn't really need an explanation. Even though I'd uncovered an error in the company's accounting procedures, I'd made an even bigger error in the process -- I made our exec. look bad when he handed out the correct report during his power breakfast meeting. It turned out that the numbers weren't so rosy that week as they'd been in previous weeks. The other company heads wanted to know why. I'm not sure what our exec. told them then, but I can't imagine it made him look good no matter how he tried to spin it.

I suppose, if the numbers had looked better using my correct spreadsheet calculations, maybe I'd have received a raise from that exec. In this particular case, and much to my surprise, the wrong answer was the right answer in his method of bookkeeping. Frustrated by this incident, I left the accounting business soon afterwards. As it later turned out, the company went belly-up years later. Looking back, I like to imagine that reason was that the company's bankers were using spreadsheets based on mathematics instead of wishful thinking. Then again, after seeing what happened with Enron, I wonder if the bankers were in on it too.

Anybody who trust spreadsheets is a fool

[crovira]

Spreadsheets are the triumph of the presentation of data over common sense.

Quite apart from the fact that some of the calculations are just plain wrong (and lack precision and are subject to systemic errors, such as rounding) the 'matrix' is often abused into serving as a data table.

Dan Bricklin may have latched onto a good thing for Apple but the unleashed idiots who used VisiCalc (and have continued through to Excel where the spreadsheet metaphor finally died,) have never had more than a clue as to what it really was.

ANYONE who trusts it to be any more than a "what-if" exercise is an IDIOT.

ANYONE who trusts it even as a "what-if" exercise without backing up the calculations with some sound math is an IDIOT.

ANYONE who uses it to store tabular data is an IDIOT.

Basically, its a fucking nuisance.

I once had to verify some calculation routine that we were using because 'the user said it was wrong.'

I spent two friggin' weeks going over code, trying it over and over again until I could prove that the code was accurate to fifteen digits on either side of the decimal place with ALL the friggin' math equations.

I finally asked my boss where the error was supposed to be. How did the user know that the calculations were wrong?

I was told to wait and he'd go and ask... The end result was that the IDIOT was actually using Excell to calculate yields on some very large and very long term bonds that they were trading.

The IDIOT was actually expecting my software to give him the same results as Excell.

Never mind that Excell is a fuckin' toy with rounding errors on nine digits (I said these were LARGE bonds) and the calculations used Newton's method of approximating integrals.

I felt like killing them all for wasting my time like that.

If they'd have told me what I was REALLY trying to prove (that Excel is a piece of shit) I could have done that in a minute just sitting down with the user and letting try things (that I knew weren't going to work) to try to prove to me that Excell was an accurate calculator.

Thank God he wasn't trying to use it to store his data. (I'd have ripped him a new ass-hole with a rusty can opener.)