@@NotKewl According to people on Twitter, quite a few people, Benford's law before the election was a decent tool and it had its uses to detect fraud. After the election, it is just a pipe dream of the sore losers and not reliable. Life comes at ya fast.
yes and was sad to find out this law doesn't work well for election fraud detecting. see description and thesis paper showing it didn't work in the past: ua-cam.com/video/vIsDjbhbADY/v-deo.html&authuser=0
@@dylanmcgriff822 Think you may have the wrong link in your comment. The ua-cam.com/video/vIsDjbhbADY/v-deo.html&authuser=0 link goes to a video just explaining the law and does not address how it applies to elections.
@@dylanmcgriff822 Update, found the paper link in video description and read it and left comment on that video. The paper was full of much rhetoric but when it finally got to a real argument it picked artificial district sizes in a simulation--which seems ridiculous if they were really interested in simulating real life. From their description of their simulation: " "Our simulations, though, consist of two types. In the first type, we simulate an election by creating 1000 districts wherein each contains the same fixed number of eligible voters. Here we run several sequences of elections where every district contains 1000 eligible voters, elections where every district contains 10,000 voters, and elections where every district contains 20,000 voters".
@@Rabcpafirm I’m taking Stat analysis next semester. Just got finished with Calc 3. (Mechanical Engineering Major) I haven’t been this excited over a math class in a while. I’m gonna be pluggin ALOT of data sets into this formula using excel once I learn more about it.
@@doombybbr thats actually a mistake. they would have to inject numbers that aren't random, they would need to inject numbers starting with 1s at greater rates than others.
@@mwatkins0590 Randomness can be skewed to favour certain patterns. There isn't just one "random" type. But it barely matters as someone can do a base 2,3,4,5,6,7,8 and 9 check and then you get caught intentionally avoiding benfords
some kind of miracle... i’m not going to go out and riot if we have to live under a Biden administration but man...his gun reform policies is OUTRAGEOUS. no online gun sales, ammo sales, or gun accessory sales: gun parts (springs, screws, pins, slings, mags, etc.)...a new age identity politics...scary stuff.
Why? Because a bunch of Republicans falsy think they can apply Benford's Law in a place where it SHOULD NOT under any circumstance be applied? Watch this video: ua-cam.com/video/etx0k1nLn78/v-deo.html
@@TomGalonska OK... Looks like Trumps data is way outta wack...But Biden's is A-OK. Sounds good? So... Umm.... Why don't we look into that? Do you see a lot of Republicans saying that if there was fraud on behalf of Trump we shouldn't look into that? What I am seeing is almost universal, the Republicans want a second look at the numbers where as the Democrats, even when the numbers show that it is Trump doing the tampering, do not want to look at the numbers and are fine with the result. Frankly, it doesn't matter what party you subscribe to, these anomalies should encourage you to look further into the counts no? Not only that, but the districts that Quick Maths provided were of Chicago where is this Forensic Accountant did all of Georgia. IT DOESN'T MATTER WHOSE SIDE YOU ARE A FANGIRL FOR! THE ANOMALIES ARE THERE ON BOTH SIDES NO? THEN LOOOOOOOOK INTO IT!!!!!!!!!!!!!!! Its not the Republicans that are having a tough time getting it this time around.
Accountants using math and never heared about statistical standard deviations. Whenever someone say "Its very complicated, it really means, he did not understand the thing"
No, you can’t use Benford’s Law to tell whether this cdc data is manipulated. For one the dataset is much too small (only 56 data points), giving random events too much weight. You can do the same analysis for the population count of the US states and will get a similar “fraudulent” looking histogram.
From a Statistician: The sample size you're using for this data analysis is way too small. I did this myself using the coronavirus cases for every county in the United States and found that the data follows Benford's Law almost perfectly. Another condition of Benford's Law that is being violated is that you have to have a data set that spans a large number of values. In this case, you have a very high preponderance of cases that fall in the Tens of Thousands and Hundreds of Thousands. I can link the data sheet if anyone is interested.
Honest question, If you have a randomly generated sample of numbers, would its conformity to benfords law depend on sample size? In other words does benfords law not hold for smaller sample sizes, even if the sample is truly randomised? Or would the conformity just not be as pronounced as a larger sample size?
Another way you know this video analysis isn't very good is that it can't figure out whether this phenomenon is called "Benford's Law" or "Benson's Law" - if you can't bother to get that right, why should anyone trust that your analysis is performed any more carefully?
To those wondering, Benfords law importantly requires values from multiple orders of magnitude inorder to be accurate (eg. 10-1000). If it happens to be that the large majority >90% happen to be in the same order of magnitude you begin to see a normal distribution
You are exactly right. Also, there is no assessment of statistical error in this analysis. I am a scientist and professor. My laboratory routinely use statistics to draw conclusions about the results of experiments. The analysis in this video is completely misleading. I stumbled upon this video, and I am deeply disturbed by the implications and the vast majority of comments. Today I made a video explaining how to properly analyze the COVID-19 data using county data which is a much larger data set. ua-cam.com/video/Y3P8A1RR5BM/v-deo.html I am happy to apply similar proper analyses to other data sets if people would like to see that.
yes and was sad to find out this law doesn't work well for election fraud detecting. see description and thesis paper showing it didn't work in the past: ua-cam.com/video/vIsDjbhbADY/v-deo.html&authuser=0
Exactly, seeing people in here think 50 states is enough is asinine. You might be able to do this with county level data as it should provide enough data points, but even then it might not be applicable to epidemiology and it’s a weak test! It can’t prove fraud. It can only reject the assumption that there is no fraud. Lotta armchair statisticians in the comments section Also love that his graph says Bensons Law
@@OxfordLlama92 This is so weird, right? The guy spends the first 5 munutes explaining Benford's Law can be used on large dataset, and then he shows an example of a tiny dataset. Anyone who has done a semester of Statistics 101 should understand that's not how it works.
Thank you for this comment! The data set is too small. Although you can use statistics on small data sets, it has to be done properly with a proper assessment of statistical error, which surely was not done here. I am a scientist and professor. My laboratory routinely use statistics to draw conclusions about the results of experiments. I stumbled upon this video, and I am deeply disturbed by the implications and the vast majority of comments. Today I made a video explaining how to properly analyze the COVID-19 data using county data which is a much larger data set. ua-cam.com/video/Y3P8A1RR5BM/v-deo.html I am happy to apply similar proper analyses to other data sets if people would like to see that.
To everyone asking for a thorough Benford’s Law analysis of the Chicago precinct election results, the mathematician Matt Parker did a thorough look on his UA-cam channel Stand-up Maths. The long and short of it is that Benford’s Law is not suitable for all data sets. Specifically, it’s a good indicator only for datasets that span multiple orders of magnitude. Consider a dataset that follows a normal (bell curve) distribution. If, for example, all of these data are one-digit numbers, then the Benford’s Law graph would be equivalent to a histogram of the data; it would look like a bell curve, which doesn’t match Benford’s Law. This doesn’t mean the data is fraudulent, it just means that it doesn’t cover a wide enough range for Benford’s Law to be applicable. A real-world example of this would be the distribution of adult heights in inches- all of the data will be two-digit numbers, so plotting all of the first digits is the same as just grouping all of the data into 10-inch bins. Legitimate data, doesn’t follow Benford’s Law. The issue with the Chicago precincts is twofold: first, 98% of the precincts had vote totals between 100 and 1000 (average of 516, stdev of 173) and second, Biden was the favorite in the majority of the precincts. This means that the vast majority of Biden vote totals were three-digit numbers, while about 30% of Trump vote totals were three-digit numbers with 70% as two- or one-digit numbers. This means that, like in the example above, the Biden graph of the data is essentially a histogram with a bin size of 100, which we would reasonably expect to follow a normal distribution, which we see. Meanwhile the Trump data significantly covers three orders of magnitude, making it a suitable dataset for Benford’s Law.
You should probably mention that Benford's law only applies to datasets that cross several orders of magnitude. A lot of people are being mislead by this with the election results. If, for example, all the values in the dataset are two digit numbers then Benford's law does NOT apply.
Trump vote totals conform rather well to Benford's law, one of the first things you'd check in trying to determine whether there was fraud. Biden's fail miserably, and, what's more, the frequencies of leading digits for him are virtually symmetric about 5: exactly what you get if these were faked by those ignorant of Benford's law, and wanting to manufacture "average" looking numbers, those that begin with 4,5, or 6
@@d1agram4 I was sad to find out this law doesn't work well for election fraud detecting. see description and thesis paper showing it didn't work in the past: ua-cam.com/video/vIsDjbhbADY/v-deo.html&authuser=0
The numbers in a phone book won't follow Benford's law. Also there are some assumptions that must be true in your data for Benford's law to be applicable.
yes and was sad to find out this law doesn't work well for election fraud detecting. see description and thesis paper showing it didn't work in the past: ua-cam.com/video/vIsDjbhbADY/v-deo.html&authuser=0
@@dylanmcgriff822 Don't be too sure. I don't have much of a head for Maths, although I did study the logarithmic way long, long ago ..LOL. But if I understood correctly, the boy says that the pattern is consistent and always works. Perhaps not in random situations like the front page of a newspaper, but if you listen to the full video, he explains why it DOES work for every fraud situation. I could be wrong in my understanding, because this bright young man did garble my brain with the complex mathematics, but I think Benford's Law will work if applied to this election situation, if only to see if fraud has actually occurred (I think it has!). In fact, I saw somewhere in this comment section, that this fellow has already been engaged by several parties concerning the US presidential election. Keeping fingers crossed that they will find what they need to find so that democracy will prevail!!
It has been applied, but the videos showing it are being censored by youtube...I WONDER WHY is the question every journalist on the planet earth should be asking right now!
@@sblbb929 Wow! Thanks for that. These graphs prove that Biden is the only one that violates Benford's Law. All the rest are more or less following the pattern. And that's just in Chicago. This proves that, YES, the Benford principle does indeed work even for elections. Thank you so much!
@Smarmy Fellow no, its not. the "maths" he's trying to use here is not apt at all for this situation. if you want, you should look up benfords law yourself and watch a vid from an actual mathematician so you understand
@@notme-ji5uo I agree to an extent, but that's not to suggest what he's explaining and uncovered isn't applicable in this case. He didn't say the reasons why the data could be misleading, only that something merits further investigation. I understand what you're saying though, due to other unforeseen circumstances (depending on the situation and circumstances) these anomalies would alter the results making the progressive curve less uniform. As he also said, this Benford's Law is a preliminary tool to look for anomalies, perhaps he should have then delved in deeper from that point on to see what else were the actual causes.
@@mengyuanxian3279 Nah, they just watched one video one the results from the Detroit area and became instant experts. Next they'll prove the USPS is fraud because zip codes don't follow Benford's law.
@@SBBurzmali Even without benford's law theres been several witness accounts from postal workers and poll workers directly corelating to some of the fradulent/compromised votes we see
I thought that too. Especially if it’s a local phone book, the numbers will have a higher likelihood of starting with a particular number due to area codes. Now if you took all the phone numbers in the US, that might be different. I wonder what he meant
yes and was sad to find out this law doesn't work well for election fraud detecting. see description and thesis paper showing it didn't work in the past: ua-cam.com/video/vIsDjbhbADY/v-deo.html&authuser=0
@@dylanmcgriff822 No, it actually works As the only detraction is that certain areas tend a little bit towards certain leading digits, but not enough that a benfords law with a large data set would fail to follow the general trend. It is only a distinction that matters if you are looking at very minor deviations from benfords law, it doesn't apply to massive deviations. You just get the curve being a little steeper or a little less steep, the general curve still exists. Not to mention, every other location in this election follows the trend. And the only ones that don't just happen to be for biden in the massive swing states.
I did a presentation of Benford's Law in my fraud class in the Masters of Accounting Degree. I would do an overlay of the presented number against the Benford's distribution.
can you tell me about fake data entry i mean how can we know the data is fake or authentic? for example if someone has 150 responses and duplicate the data and make some changes in it and make it 300 is there any way we can find out that the data of 150 responses is fake or not?
I would argue that the randomness or deviation from natural order is less when you find randomness applied non universally, in other words if everyone is cheating, the magnitude or severe-ness of the cheating is also a measurement of natural order. The regularity or uniformity with which misreporting is applied can contribute to Benfordness.
This is such a small data set that you would expect some statistical variation. Digits 1 and 2 are the biggest but in a small set, the remaining digits are basically meaningless
Benford's Law should be used as a "possible" indicator of fraud. It can lead you in some directions but it shouldn't be taken to be a definitive indicator of fraud on small datasets.
Apply this law to each county & state electoral return's and publish the results. It will be very interesting to see how each state is different It's already been proven in Chicago.
If the number 6 has the second-highest frequency, then it probably means that the data is not random or the sample size is not large enough. When our analysis comes up with the number 6 as the second-highest frequency, it’s usually telling us that the data is manipulated and probably fraudulent. Thanks for the observation Robert
I was sad to find out the Benford's Law doesn't work well for election fraud detecting. see description and thesis paper showing it didn't work in the past: ua-cam.com/video/vIsDjbhbADY/v-deo.html&authuser=0 core.ac.uk/download/pdf/206427437.pdf
I’m new to the concept of Benford’s law, but isn’t the chart at the end indicating that it actually falls pretty well in line with it? It deviates very slightly and matches up very well considering this is a pretty small sample size. The chart for physical constants deviates more than these numbers. I feel like saying these covid case numbers are indicative of fraud is quite misleading.
@@melinawright5350 For argument's sake, lets say that it is fraudulent, this benford law chart does not indicate that. It doesn't deviate that much and it finds the trend very well considering how small the sample size is.
A good video, but I gotta say that Covid data looked like it conformed pretty well with Benford's Law. There are only a few dozen data points, which is not many when you have nine slots to fill with data. Obviously, 1 is going to be the number that starts to take shape first as you add data and produce the Benford graph. 2 will be the second quickest slot to take it's proper shape. In this data, 1 occurs most frequently and 2 is already in the 2 spot. The rest of the graph deviates, but not wildly, and there are limited data points. I think the Covid data follows the rule decently well.
A question, if we take a supplier ledger account balance every end of day and plot it so its results should be same as benford law ? or suppose a cash ledger debit entries and then credit entries balance will they follow this law ? Please elaborate
If the numbers are random, you can utilize ledger account balances or cash ledger debit entries and credit entries. Remember, Bedford’s law only tells you there may be a problem; you have to do extensive forensic accounting to substantiate any anomalies
Does the example with phone numbers in a phone book work? I thought they used specific numbers for specific things which would throw it off Edit: also how many data points do you typically need before you see this technique as being significant? Obviously if you have millions of values it should be good, but is the example with 56 enough to make a judgement or would you typically need a lot more in a real world example?
Now that's some very good short example, quite helpful for getting a quick, basic grasp of how the law gets applied on practical level. You just earned another sub sir!
In covid example there are only 50 data points, why would the law hold when only so few data points? If you do a hypothesis test will that variance be significant given only 50 points?
Hi Robert, thanks for sharing this information. Probably gonna sound like a broken record here in the comment section but I am too here due to the 2020 election.
Benford's law is not a law, likewise Moore's law and Murphy's law are laws neither. They are all rules which strongly depend on the boundary conditions. So this article is a big failure.
Unfortunately not - there are major flaws in this application (e.g. it is based on just 50 numbers whereas Benford's law would need much larger datasets across several orders of magnitude to have any chance of the dataset conforming to the pattern shown in Benford's law. Check this video out for a much more insightful analysis on Benford's law. ua-cam.com/video/etx0k1nLn78/v-deo.html
The database you used for the example looks too small, 7 days' data seems to be small of sample size to be accurate. I heard it should be over something like 2000 to be more accurate, is that true?
What's scare me is, the guy said it use this for fraud and go to court with these and he doesn't understand at all the limits of this Law. Even if you don't know Benfords law you can see why this case (Covid with state) can't be use with this law. First, you must have a lot of data, like he said himself, 56 is not a lot. Second, you must have a wide range of magnitude in the data. Because if the state population is around the same (most of them) it will produce a graph that didn't follow Benfords law. Think of it, if most of the state are around 5-6-7 million and a mean % of people die of covid, like 0,01% maybe 0,011% or 0,009% it will produce a lot of data starting with the same number, and It will not necessarily be the number 1. And it can be prove with the same example but with the population in the states. I did the graph with the population by state and strangely the graph is almost the same as the one he show for the covid by state (5-6-7 are like a bump in the graph). So if I follow his reasoning, the population by state have been manipulated. That's not an absolute law, a lot of things can introduce a biais in it, and data specialist should be able to analyse that and conclude that the law can't be use in this case to prove, but more often to investigate on a case. He said that it can be use the see if data is random or manipulated, that is false. If data is a near true random as a computer do random data, it's not gonna follow the Benfords law, all graph bar will be at the same level. I don't said that the data are correct, but this Law can't be use in this case. This guy should'n go to court or represent itself as a specialist. If I was the defendant and this guy showed up, i'll show this video to discredit him.
Regardless of political affiliation, the possibility of fraud in a free and fair society should disgust and concern *everyone*. Unless we want to become a third world nation, this must be taken seriously and thoroughly investigated. Thank you for such a fantastic explanation and analysis.
We need to spread this as far as we can. I learned of Benford's Law from Louder with Crowder. Hopefully the President can learn of this fraud detection technique.
Bensford's law only applies when there is a large range in the sizes of the data sets. en.wikipedia.org/wiki/Benford%27s_law It would have been nice if you had mentioned it. There may also be a misunderstanding of the definition of "law." A law is "a generally valid description, usually mathematical." In engineering, for example, Hooke's law is not valid for all materials over all stresses.
Applying this test outside of financials is tricky. It works well with general ledger items because the prices come from unrelated external sources. It's great for accounting. When you're looking at something from a single measured source, like these COVID results, the data needs a spread of many orders of magnitude for the test to be useful. Try this: Measure the height, in inches, of everybody you know and run it through this test. Why are 6 and 7 dominating everything? Did you manipulate the data? Of course not.
I'd recommend to all viewers to check out the video by Matt Parker on his Stand-up Maths channel where he explains Benford's Law and the US 2020 election results in greater detail. Gives good insight and will help put to rest any conspiracy theories as far as at least Benford's law is concerned. The apparent anomalies in the election data when Benford's Law is applied are discussed and proven to be as expected given the constraints of the system. If there is some sort of conspiracy, this is provably not it.
Can Benford's law be assistive in investigating fraud cases where fraudsters misused the Covid relief funds? Is there a pattern observed in cases already investigated?
@@Rabcpafirm I recently read an article in NewYork times. It said that the government released $5 Trillion in relief funds. Billions of dollars from these grants were misused by fraudsters …. There are thousands of such fraud cases that the investigators are looking into. I was wondering if Benford law could be helpful there.
I swear I never typed ‘benfords law’ into my phone or computer just listened to podcasts talking about it and boom this is recommended, sure could just be viewed many times lately so UA-cam recommends it okay. Still but weird. What’s even crazier is this was made before the 2020 election votes. Talk about good timing to make this UA-cam video!
Do you not do an actual test on this data? You should compare it to the expected frequencies given by benfords law sum((observed # of specific digit-total obs*expected frequency)/total*expected frequency) The expected distribution of this is chi-squared so at 0.05 we have a 16.9 critical value. Simply showing this graph without the underlying test is wrong.
Yes! I fair portion of my job includes plotting and interpreting data sets in excel. Very cool, and I am not surprised in the least that hospitals would over report covid data since they are given money for every positively identified covid patient in the hospital system. I also work for a clinic, so I take special interest in these things. The money is supposed to go to helping the hosptials take care of covid positive patients, but if you report extra, that means some extra cash to help support other needs of the hospital. I wouldn't be surprised this is what is going. It was a very poor system with too much incentive for fraud, and not enough measures in place to combat that.
Unfortunately, this forensic accountant has applied Benford's law in a majorly flawed way in this video. Benford's law applies in 'large' datasets in certain situations. Here it is applied to a small dataset which is one reason it has minimal chance of getting close to the Benford's law pattern. It certainly is not proof that data has been manipulated.Check this video for a more accurate description of Benford's law ua-cam.com/video/etx0k1nLn78/v-deo.html
Before anyone gets too excited, check out this video ua-cam.com/video/etx0k1nLn78/v-deo.html which explains why Benford's Law is not an appropriate metric for most data at the precinct level. TL;DR: The precincts are defined to be roughly equal sizes all around 500 - 1000 voters. If a candidate got about 70% of the vote on average you would expect all of his/her vote counts to be 3 digit numbers with and average around 500-600. This means Benford would not tell anything much because it only works when spanning multiple orders of magnitude. Benford's law works in general because a random sample of an exponential distribution is going to get more data from the low end (i.e. 1s and 2s). But a sample that is constructed to be from data of equivalent sized populations is not going to be an exponential distribution.
How does the size of the data set affect the fit of that logarithmic curve? Simple chance must be involved for small data sets. After all, a Data set of one value will have a 100% spike on one digit. Is there a rule for how many numbers a data set must have before you rely on this method?
Phone numbers do not follow the law. In Finland for example all numbers begin with 0. Benford's law applies in figures were the variation is high there is no clear upper limit, like sums that can be just a few dollars to thousands of dollars. It does not apply to cases where the variation is small and in a limited range like human heights.
Branford’s law is an interesting phenomenon and some years ago Steven W Smith in his book “The Scientist and Engineer’s Guide to Digital Signal Processing” covered this topic, which I recommend to everyone. My comment for you is to apply the Chi-square test to the counts of the 9 leading numbers remembering to group if a cell has less than 5 counts. This will indicate if the distribution of the data deviates from the theoretical one. As you and your subscribers probably know this method can and has been applied to election data.
I applied Benfords law to a variety of datasets that I had used in my college days and not one follows it. The datasets ranged from modules relating to finance, economics, marketing and 1 HR subject. I performed most of them with high integrity and caution yet not one of them follows this law. What could be some of the possible explanations for this?
Branford’s law requires a data set of numbers that span large orders of magnitude. For example Benford’s law will not work for the height of males in metres in a given population because almost all heights will begin with the number 1 and only a tiny number will begin with 2 and there will be no heights beginning with a 3 or higher. There may be dwarfs that are shorter than 1m - not sure. The point is that you need to specify what parameter and range of measurement is applicable to the data set before you can ascertain whether Benfords law applies or is a useful analytical tool to deploy What actual numbers or quantities did you choose? Pr(d) = log(1+1/d) Where Pr(d) is the probability of a particular leading digit, d, appearing in the data set. If you substitute d=1 you get Pr(1) = log(2) = 0.30103 or 30.1% Pr(2) = log(3/2) = 0.1761 or 17.61% and so on..
Well we are actually detecting an anomaly that requires investigation of the underlying data. To say that it will detect a lie is in itself a falsehood.
Great easy to understand video. And no surprise that different states are reporting cases in different ways, not to mention the inaccuracy of the tests. One common mistake is to assume the underlying data for each data set member is incorrect when in fact, the results here only show the randomness or lack thereof for the data set as presented. This example illustrates that quite well.
Mr. Bonavito, for data to look suspicious, what ratio between observed and expected should I use as threshold? For instance, if +/- 25% is the fraud threshold, and the "1"s are outside of +/- 25% of the expected 0.3 (higher than 0.0375 or lower than 0.0225), then I would view the data as suspicious.
You had a keen observation. Mr Bonavito used "eyeball" analysis. The histogram did not look Benford was his conclusion. You want a more quantitative analysis. There are many, many ways to measure Benford quantitatively. The most common is the Chi Square test. You asked for a ratio that deviates from the expected value. You are on the right track because it is an excellent test but unfortunately there is no simple ratio. You would have to calculate the Z score for all 9 digits. To be 95% confident that Mr.Bonavito's example DOES NOT follow Benford the Z score should be greater than 2.56(because 9 digits are being tested the Bonferoni adjustment comes in to play). I calculated the Z score for 3 digits and not any was over 2.56. An example to calculate the Z score for leading digit 1 is: (.41-.30)/((.30x.70)/56)^.5= 1.8 .41 is the actual percentage .30 is the expected percentage .70 is 1 minus the expected percentage 56 is the sample size ^.5 is the square root All 9 digits can be tested that way using each digits own specific actual and expected percentages.
So the assumption is that because the distribution does not match the expected outcome, there is fraud likely involved. But with Benford's law being a statistical instrument, it is tied to statistical annomylies that increase massivly for small sample sizes (anything below 1000 data points id argue is a small sample size). So at what point do you say "ok this is just not enough data points for me to make a judgement call on it" instead of "here is likely fraud in play" ?
One Month Ago... Bruh, he drops the smoking gun a month before it happens. This man is a psychic!
UA-cam changed the date
Yeah it’s not like Benford’s law existed before the election or was ever used
@@NotKewl According to people on Twitter, quite a few people, Benford's law before the election was a decent tool and it had its uses to detect fraud. After the election, it is just a pipe dream of the sore losers and not reliable.
Life comes at ya fast.
It was expected that the dems would cheat for a long time now
It was used to catch Bernie Madoff
Upvote if you are here because of the 2020 election “results”.
yes and was sad to find out this law doesn't work well for election fraud detecting. see description and thesis paper showing it didn't work in the past: ua-cam.com/video/vIsDjbhbADY/v-deo.html&authuser=0
@@dylanmcgriff822 Think you may have the wrong link in your comment. The ua-cam.com/video/vIsDjbhbADY/v-deo.html&authuser=0 link goes to a video just explaining the law and does not address how it applies to elections.
@@reasonforge9997 Yes, read his 2020 election update in his discriptions below. He references this paper: core.ac.uk/download/pdf/206427437.pdf
@@dylanmcgriff822 Update, found the paper link in video description and read it and left comment on that video. The paper was full of much rhetoric but when it finally got to a real argument it picked artificial district sizes in a simulation--which seems ridiculous if they were really interested in simulating real life. From their description of their simulation: " "Our
simulations, though, consist of two types. In the first type, we simulate an election by creating 1000
districts wherein each contains the same fixed number of eligible voters. Here we run several sequences
of elections where every district contains 1000 eligible voters, elections where every district contains
10,000 voters, and elections where every district contains 20,000 voters".
This helps prove voter fraud!
A CPA: The hero we need
Have no fear, New Jersey Forensic Accountant is here!
Did a math degree but got bored of it, maybe this is the motivation needed to set out a path to forensic accounting!
@@Rabcpafirm this is what he going to use in court?
@@atlnlgga3844 probably not, but gives people the hint that something does not going well
@@Rabcpafirm
I’m taking Stat analysis next semester. Just got finished with Calc 3. (Mechanical Engineering Major)
I haven’t been this excited over a math class in a while.
I’m gonna be pluggin ALOT of data sets into this formula using excel once I learn more about it.
I don't think Biden/Harris and Dr Fauci knew about this law.
even if they did, they would need to inject a random number of votes into every state in order to hide from it.
@@doombybbr that's not practical when trying to add the minimum number of votes to win but enough to not trigger a recount in each state.
@@doombybbr thats actually a mistake. they would have to inject numbers that aren't random, they would need to inject numbers starting with 1s at greater rates than others.
@@mwatkins0590 Randomness can be skewed to favour certain patterns. There isn't just one "random" type.
But it barely matters as someone can do a base 2,3,4,5,6,7,8 and 9 check and then you get caught intentionally avoiding benfords
If you actually care enough about this to watch a 2 minute intro, here you go: ua-cam.com/video/etx0k1nLn78/v-deo.html
Robert my man, get ready for a Red wave of Subs. 🙏🏾
Unless he gets removed for being fake news
@@PRODAt3 ARCHIVE EVERYTHING!
some kind of miracle... i’m not going to go out and riot if we have to live under a Biden administration but man...his gun reform policies is OUTRAGEOUS. no online gun sales, ammo sales, or gun accessory sales: gun parts (springs, screws, pins, slings, mags, etc.)...a new age identity politics...scary stuff.
Why? Because a bunch of Republicans falsy think they can apply Benford's Law in a place where it SHOULD NOT under any circumstance be applied? Watch this video: ua-cam.com/video/etx0k1nLn78/v-deo.html
@@TomGalonska OK... Looks like Trumps data is way outta wack...But Biden's is A-OK. Sounds good? So... Umm.... Why don't we look into that?
Do you see a lot of Republicans saying that if there was fraud on behalf of Trump we shouldn't look into that?
What I am seeing is almost universal, the Republicans want a second look at the numbers where as the Democrats, even when the numbers show that it is Trump doing the tampering, do not want to look at the numbers and are fine with the result.
Frankly, it doesn't matter what party you subscribe to, these anomalies should encourage you to look further into the counts no?
Not only that, but the districts that Quick Maths provided were of Chicago where is this Forensic Accountant did all of Georgia.
IT DOESN'T MATTER WHOSE SIDE YOU ARE A FANGIRL FOR! THE ANOMALIES ARE THERE ON BOTH SIDES NO? THEN LOOOOOOOOK INTO IT!!!!!!!!!!!!!!!
Its not the Republicans that are having a tough time getting it this time around.
Accountants using math and never heared about statistical standard deviations. Whenever someone say "Its very complicated, it really means, he did not understand the thing"
No, you can’t use Benford’s Law to tell whether this cdc data is manipulated. For one the dataset is much too small (only 56 data points), giving random events too much weight.
You can do the same analysis for the population count of the US states and will get a similar “fraudulent” looking histogram.
From a Statistician:
The sample size you're using for this data analysis is way too small. I did this myself using the coronavirus cases for every county in the United States and found that the data follows Benford's Law almost perfectly. Another condition of Benford's Law that is being violated is that you have to have a data set that spans a large number of values. In this case, you have a very high preponderance of cases that fall in the Tens of Thousands and Hundreds of Thousands. I can link the data sheet if anyone is interested.
Honest question, If you have a randomly generated sample of numbers, would its conformity to benfords law depend on sample size? In other words does benfords law not hold for smaller sample sizes, even if the sample is truly randomised? Or would the conformity just not be as pronounced as a larger sample size?
hello, can i have a link?
awesome
@@gulzarkundan4470 it doesn't work for purely randomly generated numbers, that's why frauds don't follow it
Another way you know this video analysis isn't very good is that it can't figure out whether this phenomenon is called "Benford's Law" or "Benson's Law" - if you can't bother to get that right, why should anyone trust that your analysis is performed any more carefully?
To those wondering, Benfords law importantly requires values from multiple orders of magnitude inorder to be accurate (eg. 10-1000). If it happens to be that the large majority >90% happen to be in the same order of magnitude you begin to see a normal distribution
You are exactly right. Also, there is no assessment of statistical error in this analysis. I am a scientist and professor. My laboratory routinely use statistics to draw conclusions about the results of experiments. The analysis in this video is completely misleading. I stumbled upon this video, and I am deeply disturbed by the implications and the vast majority of comments. Today I made a video explaining how to properly analyze the COVID-19 data using county data which is a much larger data set. ua-cam.com/video/Y3P8A1RR5BM/v-deo.html
I am happy to apply similar proper analyses to other data sets if people would like to see that.
Do a video on the election using Benfords Law. Thank you.
yes and was sad to find out this law doesn't work well for election fraud detecting. see description and thesis paper showing it didn't work in the past: ua-cam.com/video/vIsDjbhbADY/v-deo.html&authuser=0
@@dylanmcgriff822 paid shill
Dylan McGriff probably edited the wikipedia page to say that too, thats why he links to a random video about the law.
Check out his new video: ua-cam.com/video/DoF3WS42w3M/v-deo.html
@@r-gart The guy made a vid on how Benfords law does not work for elections in 2011. He's not a paid shill. You;re just a moron
Branford’s law is known to not work well with data sets with 500 or fewer transactions.
Exactly, seeing people in here think 50 states is enough is asinine. You might be able to do this with county level data as it should provide enough data points, but even then it might not be applicable to epidemiology and it’s a weak test! It can’t prove fraud. It can only reject the assumption that there is no fraud.
Lotta armchair statisticians in the comments section
Also love that his graph says Bensons Law
@@OxfordLlama92 This is so weird, right? The guy spends the first 5 munutes explaining Benford's Law can be used on large dataset, and then he shows an example of a tiny dataset. Anyone who has done a semester of Statistics 101 should understand that's not how it works.
Thank you for this comment! The data set is too small. Although you can use statistics on small data sets, it has to be done properly with a proper assessment of statistical error, which surely was not done here. I am a scientist and professor. My laboratory routinely use statistics to draw conclusions about the results of experiments. I stumbled upon this video, and I am deeply disturbed by the implications and the vast majority of comments. Today I made a video explaining how to properly analyze the COVID-19 data using county data which is a much larger data set. ua-cam.com/video/Y3P8A1RR5BM/v-deo.html
I am happy to apply similar proper analyses to other data sets if people would like to see that.
I bet you had no idea this video was going to blow up due to the elections.
To everyone asking for a thorough Benford’s Law analysis of the Chicago precinct election results, the mathematician Matt Parker did a thorough look on his UA-cam channel Stand-up Maths.
The long and short of it is that Benford’s Law is not suitable for all data sets. Specifically, it’s a good indicator only for datasets that span multiple orders of magnitude. Consider a dataset that follows a normal (bell curve) distribution. If, for example, all of these data are one-digit numbers, then the Benford’s Law graph would be equivalent to a histogram of the data; it would look like a bell curve, which doesn’t match Benford’s Law. This doesn’t mean the data is fraudulent, it just means that it doesn’t cover a wide enough range for Benford’s Law to be applicable. A real-world example of this would be the distribution of adult heights in inches- all of the data will be two-digit numbers, so plotting all of the first digits is the same as just grouping all of the data into 10-inch bins. Legitimate data, doesn’t follow Benford’s Law.
The issue with the Chicago precincts is twofold: first, 98% of the precincts had vote totals between 100 and 1000 (average of 516, stdev of 173) and second, Biden was the favorite in the majority of the precincts. This means that the vast majority of Biden vote totals were three-digit numbers, while about 30% of Trump vote totals were three-digit numbers with 70% as two- or one-digit numbers. This means that, like in the example above, the Biden graph of the data is essentially a histogram with a bin size of 100, which we would reasonably expect to follow a normal distribution, which we see. Meanwhile the Trump data significantly covers three orders of magnitude, making it a suitable dataset for Benford’s Law.
You should probably mention that Benford's law only applies to datasets that cross several orders of magnitude. A lot of people are being mislead by this with the election results. If, for example, all the values in the dataset are two digit numbers then Benford's law does NOT apply.
AND only if the dataset is big enough.
And even if the criteria are met and there is a violation, it's an indicator, not proof
Wow I stumbled upon this. Great presentation by the way, I’m wondering, can you apply Benfords Law to the results Of the 2020 Presidential Election?
Yes, and how has this been applied to other elections in failed democratic elections?
Thanks Pb X, really glad you enjoyed the video. Yes, unfortunately, the 2020 election results will have similar issues when we apply Benford’s law.
Trump vote totals conform rather well to Benford's law, one of the first things you'd check in trying to determine whether there was fraud. Biden's fail miserably, and, what's more, the frequencies of leading digits for him are virtually symmetric about 5: exactly what you get if these were faked by those ignorant of Benford's law, and wanting to manufacture "average" looking numbers, those that begin with 4,5, or 6
@@d1agram4 I was sad to find out this law doesn't work well for election fraud detecting. see description and thesis paper showing it didn't work in the past: ua-cam.com/video/vIsDjbhbADY/v-deo.html&authuser=0
@@dylanmcgriff822 shill
The numbers in a phone book won't follow Benford's law. Also there are some assumptions that must be true in your data for Benford's law to be applicable.
I've subscribed because I want to see a video of Benford's law applied to this election.
yes and was sad to find out this law doesn't work well for election fraud detecting. see description and thesis paper showing it didn't work in the past: ua-cam.com/video/vIsDjbhbADY/v-deo.html&authuser=0
@@dylanmcgriff822 Don't be too sure. I don't have much of a head for Maths, although I did study the logarithmic way long, long ago ..LOL. But if I understood correctly, the boy says that the pattern is consistent and always works. Perhaps not in random situations like the front page of a newspaper, but if you listen to the full video, he explains why it DOES work for every fraud situation. I could be wrong in my understanding, because this bright young man did garble my brain with the complex mathematics, but I think Benford's Law will work if applied to this election situation, if only to see if fraud has actually occurred (I think it has!). In fact, I saw somewhere in this comment section, that this fellow has already been engaged by several parties concerning the US presidential election. Keeping fingers crossed that they will find what they need to find so that democracy will prevail!!
It has been applied, but the videos showing it are being censored by youtube...I WONDER WHY is the question every journalist on the planet earth should be asking right now!
pbs.twimg.com/media/EmLFD_7UYAAvvzJ?format=jpg B_den's vote results violate Benford's Law
@@sblbb929 Wow! Thanks for that. These graphs prove that Biden is the only one that violates Benford's Law. All the rest are more or less following the pattern. And that's just in Chicago. This proves that, YES, the Benford principle does indeed work even for elections. Thank you so much!
I'm amazed that youtube would recommend this. The machine AI must not realize this goes against their narrative.
@Smarmy Fellow no, its not. the "maths" he's trying to use here is not apt at all for this situation. if you want, you should look up benfords law yourself and watch a vid from an actual mathematician so you understand
@@notme-ji5uo I agree to an extent, but that's not to suggest what he's explaining and uncovered isn't applicable in this case. He didn't say the reasons why the data could be misleading, only that something merits further investigation. I understand what you're saying though, due to other unforeseen circumstances (depending on the situation and circumstances) these anomalies would alter the results making the progressive curve less uniform. As he also said, this Benford's Law is a preliminary tool to look for anomalies, perhaps he should have then delved in deeper from that point on to see what else were the actual causes.
Who's here after the 2020 us election?? Haha
If you actually care enough about this to watch a 2 minute intro, here you go: ua-cam.com/video/etx0k1nLn78/v-deo.html
Me
Everyone lol
@@thelastengineer2315 me 2....
That is EXACTLY why I'm here!
I’ll be taking this to court
that awkward moment when bidens election results break this law, and only in the swing states lol
Are you sure it's only in the swing states? Has anybody ever checked the election results in CA and NY against the Benford law?
@@mengyuanxian3279 Nah, they just watched one video one the results from the Detroit area and became instant experts. Next they'll prove the USPS is fraud because zip codes don't follow Benford's law.
@@SBBurzmali How's that CNN internship coming along ? I see your already falling in line ... you'll do just fine
@@chadbitcoin5085 lol, have you woven CNN into this conspiracy to steal the election or that just the go to attack these days?
@@SBBurzmali Even without benford's law theres been several witness accounts from postal workers and poll workers directly corelating to some of the fradulent/compromised votes we see
At 2:50 Not sure it will work on the numbers of a phone book as they are not naturally occurring counts but may be artificially distributed.
I thought that too. Especially if it’s a local phone book, the numbers will have a higher likelihood of starting with a particular number due to area codes. Now if you took all the phone numbers in the US, that might be different. I wonder what he meant
Mr. Bonavito, please do a video on this election in relations to Benford's Law.
yes and was sad to find out this law doesn't work well for election fraud detecting. see description and thesis paper showing it didn't work in the past: ua-cam.com/video/vIsDjbhbADY/v-deo.html&authuser=0
@@dylanmcgriff822 American used it to detect fraud in the 2009 Iranian election
@@dylanmcgriff822 No, it actually works
As the only detraction is that certain areas tend a little bit towards certain leading digits, but not enough that a benfords law with a large data set would fail to follow the general trend. It is only a distinction that matters if you are looking at very minor deviations from benfords law, it doesn't apply to massive deviations.
You just get the curve being a little steeper or a little less steep, the general curve still exists.
Not to mention, every other location in this election follows the trend. And the only ones that don't just happen to be for biden in the massive swing states.
@@dylanmcgriff822 Benford law still applies. It's just not as steep. It's not supposed to have such massive deviations as Biden's does.
I did a presentation of Benford's Law in my fraud class in the Masters of Accounting Degree.
I would do an overlay of the presented number against the Benford's distribution.
can you tell me about fake data entry i mean how can we know the data is fake or authentic? for example if someone has 150 responses and duplicate the data and make some changes in it and make it 300 is there any way we can find out that the data of 150 responses is fake or not?
The UA-cam algorithm is trying to tell us something but I wonder what it could be hmmm... 🤔🤔🤔🤔🤔🤔🤔
I would argue that the randomness or deviation from natural order is less when you find randomness applied non universally, in other words if everyone is cheating, the magnitude or severe-ness of the cheating is also a measurement of natural order. The regularity or uniformity with which misreporting is applied can contribute to Benfordness.
Good point, all forensic procedures have some weak aspects
Sir, I would like to thank you so very much for laying this out and doing so much good for our republic. God bless!
By misleading? Only 56 datapoints? He is a fraud.
This is such a small data set that you would expect some statistical variation. Digits 1 and 2 are the biggest but in a small set, the remaining digits are basically meaningless
I got banned from twitter for 12 hours for sharing a tweet about Benford's Law
Come to gab.
The right man at the right place at the right time
Can you do a data sheet with the US election 2020? That would be interesting.
github.com/cjph8914/2020_benfords
@@jesswilson6012 thanks dude, interesting
Benford's Law should be used as a "possible" indicator of fraud. It can lead you in some directions but it shouldn't be taken to be a definitive indicator of fraud on small datasets.
True, forensic audit would be needed
Apply this law to each county & state electoral return's and publish the results. It will be very interesting to see how each state is different
It's already been proven in Chicago.
Fascinating and informative. Great new tool to bang against an HOA if we can only get the financial data! Thx
It is a great tool, use it all the time in our forensic practice.
This video is BS. The dataset is way too small to claim anything.
@@AlcyonEldara what size data set is required to provide tangible results?
running befords law on a data set of
The hero America needs, the hero America deserves.
ive noticed that the number 6 is usually the number that is second highest on the chart when benfords laws doesn't work on the data. why is that?
or doesn't show true benfords law curve. sorry not sure if that make sense.
If the number 6 has the second-highest frequency, then it probably means that the data is not random or the sample size is not large enough. When our analysis comes up with the number 6 as the second-highest frequency, it’s usually telling us that the data is manipulated and probably fraudulent.
Thanks for the observation
Robert
@@Rabcpafirm awesome thanks for the reply
I was sad to find out the Benford's Law doesn't work well for election fraud detecting. see description and thesis paper showing it didn't work in the past: ua-cam.com/video/vIsDjbhbADY/v-deo.html&authuser=0 core.ac.uk/download/pdf/206427437.pdf
I’m new to the concept of Benford’s law, but isn’t the chart at the end indicating that it actually falls pretty well in line with it? It deviates very slightly and matches up very well considering this is a pretty small sample size. The chart for physical constants deviates more than these numbers. I feel like saying these covid case numbers are indicative of fraud is quite misleading.
They are fraudulent. Everybody knows that.
@@melinawright5350 For argument's sake, lets say that it is fraudulent, this benford law chart does not indicate that. It doesn't deviate that much and it finds the trend very well considering how small the sample size is.
A good video, but I gotta say that Covid data looked like it conformed pretty well with Benford's Law. There are only a few dozen data points, which is not many when you have nine slots to fill with data. Obviously, 1 is going to be the number that starts to take shape first as you add data and produce the Benford graph. 2 will be the second quickest slot to take it's proper shape. In this data, 1 occurs most frequently and 2 is already in the 2 spot. The rest of the graph deviates, but not wildly, and there are limited data points.
I think the Covid data follows the rule decently well.
A question, if we take a supplier ledger account balance every end of day and plot it so its results should be same as benford law ? or suppose a cash ledger debit entries and then credit entries balance will they follow this law ? Please elaborate
If the numbers are random, you can utilize ledger account balances or cash ledger debit entries and credit entries. Remember, Bedford’s law only tells you there may be a problem; you have to do extensive forensic accounting to substantiate any anomalies
Does the example with phone numbers in a phone book work? I thought they used specific numbers for specific things which would throw it off
Edit: also how many data points do you typically need before you see this technique as being significant? Obviously if you have millions of values it should be good, but is the example with 56 enough to make a judgement or would you typically need a lot more in a real world example?
Now that's some very good short example, quite helpful for getting a quick, basic grasp of how the law gets applied on practical level. You just earned another sub sir!
Like uncle Joe said "follow the science."
In covid example there are only 50 data points, why would the law hold when only so few data points?
If you do a hypothesis test will that variance be significant given only 50 points?
Hi Robert, thanks for sharing this information.
Probably gonna sound like a broken record here in the comment section but I am too here due to the 2020 election.
Good presentation. One last step to statistically decide if the number set conserves to Benford' s law is doing Chi Square test
Benford's law is not a law, likewise Moore's law and Murphy's law are laws neither.
They are all rules which strongly depend on the boundary conditions.
So this article is a big failure.
This is a very fascinating subject
cool
A practical example of how to apply Benford's Law! Great job!
Unfortunately not - there are major flaws in this application (e.g. it is based on just 50 numbers whereas Benford's law would need much larger datasets across several orders of magnitude to have any chance of the dataset conforming to the pattern shown in Benford's law. Check this video out for a much more insightful analysis on Benford's law. ua-cam.com/video/etx0k1nLn78/v-deo.html
My man you are about to get a lot of views once people start hearing about Benford’s Law
This is amazing!
The database you used for the example looks too small, 7 days' data seems to be small of sample size to be accurate. I heard it should be over something like 2000 to be more accurate, is that true?
Your dataset is very small.
What's scare me is, the guy said it use this for fraud and go to court with these and he doesn't understand at all the limits of this Law. Even if you don't know Benfords law you can see why this case (Covid with state) can't be use with this law. First, you must have a lot of data, like he said himself, 56 is not a lot. Second, you must have a wide range of magnitude in the data. Because if the state population is around the same (most of them) it will produce a graph that didn't follow Benfords law. Think of it, if most of the state are around 5-6-7 million and a mean % of people die of covid, like 0,01% maybe 0,011% or 0,009% it will produce a lot of data starting with the same number, and It will not necessarily be the number 1. And it can be prove with the same example but with the population in the states. I did the graph with the population by state and strangely the graph is almost the same as the one he show for the covid by state (5-6-7 are like a bump in the graph). So if I follow his reasoning, the population by state have been manipulated. That's not an absolute law, a lot of things can introduce a biais in it, and data specialist should be able to analyse that and conclude that the law can't be use in this case to prove, but more often to investigate on a case. He said that it can be use the see if data is random or manipulated, that is false. If data is a near true random as a computer do random data, it's not gonna follow the Benfords law, all graph bar will be at the same level. I don't said that the data are correct, but this Law can't be use in this case. This guy should'n go to court or represent itself as a specialist. If I was the defendant and this guy showed up, i'll show this video to discredit him.
Thanks, some good points
Regardless of political affiliation, the possibility of fraud in a free and fair society should disgust and concern *everyone*. Unless we want to become a third world nation, this must be taken seriously and thoroughly investigated. Thank you for such a fantastic explanation and analysis.
We need to spread this as far as we can. I learned of Benford's Law from Louder with Crowder. Hopefully the President can learn of this fraud detection technique.
Dont worry, the guy who made this video actually commented a few days ago he was contacted to help with the investigations.
LOL We are all really here for the election "results"
Bensford's law only applies when there is a large range in the sizes of the data sets. en.wikipedia.org/wiki/Benford%27s_law It would have been nice if you had mentioned it. There may also be a misunderstanding of the definition of "law." A law is "a generally valid description, usually mathematical." In engineering, for example, Hooke's law is not valid for all materials over all stresses.
It should be renamed to Biden's Law
I'm an insurance investigator; great content!
Thanks Jason
Please save Trump!
Have no fear, New Jersey Forensic Accountant is here!
Where in your Excel sheet did you use the "log" portion of the law ?
Election results
Applying this test outside of financials is tricky. It works well with general ledger items because the prices come from unrelated external sources. It's great for accounting.
When you're looking at something from a single measured source, like these COVID results, the data needs a spread of many orders of magnitude for the test to be useful.
Try this:
Measure the height, in inches, of everybody you know and run it through this test. Why are 6 and 7 dominating everything? Did you manipulate the data? Of course not.
"Biden's Law"
Im seeing a lot of criticism saying that Benfords law is not applicable to the election results. Its only applicable for accounting....
Everyone on the left rn: "im gonna pretend i didn't see that"
I'd recommend to all viewers to check out the video by Matt Parker on his Stand-up Maths channel where he explains Benford's Law and the US 2020 election results in greater detail. Gives good insight and will help put to rest any conspiracy theories as far as at least Benford's law is concerned. The apparent anomalies in the election data when Benford's Law is applied are discussed and proven to be as expected given the constraints of the system.
If there is some sort of conspiracy, this is provably not it.
Little did he know when he made this video, he would be hailed as a hero by Americans across the country.
"The logorithms are way out of wack". Im getting that tattooed
Hilarious
Can Benford's law be assistive in investigating fraud cases where fraudsters misused the Covid relief funds? Is there a pattern observed in cases already investigated?
Yes, we used it on vaccine data
@@Rabcpafirm I recently read an article in NewYork times. It said that the government released $5 Trillion in relief funds. Billions of dollars from these grants were misused by fraudsters …. There are thousands of such fraud cases that the investigators are looking into. I was wondering if Benford law could be helpful there.
@@AditiJainSyal Yes, we have used Benford's Law to focus are fraud selection targets
Would you consider doing a benford's law video for New York's COVID-19 data?
Our friend told us about Standard deviation many times. (4639, 4247, 3896, 3727, 3439, 2567, 2461 and many more)
#SloppyPoppy
Can you please show an example of a actual data set that has no fraud and the graph stating that?
I swear I never typed ‘benfords law’ into my phone or computer just listened to podcasts talking about it and boom this is recommended, sure could just be viewed many times lately so UA-cam recommends it okay. Still but weird. What’s even crazier is this was made before the 2020 election votes. Talk about good timing to make this UA-cam video!
It just hit me. I understand it now.
Nice
My lawyer told me not to comment but how to fix tax fraud can be also useful using the law I'm certainly not doing that
Cool
Do you not do an actual test on this data? You should compare it to the expected frequencies given by benfords law sum((observed # of specific digit-total obs*expected frequency)/total*expected frequency) The expected distribution of this is chi-squared so at 0.05 we have a 16.9 critical value. Simply showing this graph without the underlying test is wrong.
Yes! I fair portion of my job includes plotting and interpreting data sets in excel. Very cool, and I am not surprised in the least that hospitals would over report covid data since they are given money for every positively identified covid patient in the hospital system. I also work for a clinic, so I take special interest in these things. The money is supposed to go to helping the hosptials take care of covid positive patients, but if you report extra, that means some extra cash to help support other needs of the hospital. I wouldn't be surprised this is what is going. It was a very poor system with too much incentive for fraud, and not enough measures in place to combat that.
Unfortunately, this forensic accountant has applied Benford's law in a majorly flawed way in this video. Benford's law applies in 'large' datasets in certain situations. Here it is applied to a small dataset which is one reason it has minimal chance of getting close to the Benford's law pattern. It certainly is not proof that data has been manipulated.Check this video for a more accurate description of Benford's law ua-cam.com/video/etx0k1nLn78/v-deo.html
Before anyone gets too excited, check out this video
ua-cam.com/video/etx0k1nLn78/v-deo.html
which explains why Benford's Law is not an appropriate metric for most data at the precinct level. TL;DR: The precincts are defined to be roughly equal sizes all around 500 - 1000 voters. If a candidate got about 70% of the vote on average you would expect all of his/her vote counts to be 3 digit numbers with and average around 500-600. This means Benford would not tell anything much because it only works when spanning multiple orders of magnitude. Benford's law works in general because a random sample of an exponential distribution is going to get more data from the low end (i.e. 1s and 2s). But a sample that is constructed to be from data of equivalent sized populations is not going to be an exponential distribution.
Benford's Law... "Covid 9" GUILTY!
Nice
I never thought I'd say accounting was fun, but anything forensic is interesting.
How does the size of the data set affect the fit of that logarithmic curve? Simple chance must be involved for small data sets. After all, a Data set of one value will have a 100% spike on one digit. Is there a rule for how many numbers a data set must have before you rely on this method?
What happens when you cheack Benfords law on the 2019 flew season database?
I mean. Here we are a month later. This man gave us the clear guide we need.
Phone numbers do not follow the law. In Finland for example all numbers begin with 0. Benford's law applies in figures were the variation is high there is no clear upper limit, like sums that can be just a few dollars to thousands of dollars. It does not apply to cases where the variation is small and in a limited range like human heights.
This is brilliant! Thanks for sharing Benford's Law. It can be applied so many places.
But keep in mind Benford's law only applies to big datasets (not 56 points LOL) and orders of magnitude.
Branford’s law is an interesting phenomenon and some years ago Steven W Smith in his book “The Scientist and Engineer’s Guide to Digital Signal Processing” covered this topic, which I recommend to everyone. My comment for you is to apply the Chi-square test to the counts of the 9 leading numbers remembering to group if a cell has less than 5 counts. This will indicate if the distribution of the data deviates from the theoretical one. As you and your subscribers probably know this method can and has been applied to election data.
I applied Benfords law to a variety of datasets that I had used in my college days and not one follows it.
The datasets ranged from modules relating to finance, economics, marketing and 1 HR subject.
I performed most of them with high integrity and caution yet not one of them follows this law.
What could be some of the possible explanations for this?
Sir, this video is about to get a LOT of views as people look up Benford's Law -- and your graph has *Benson's* Law!
Benson's Law - The butler is the funniest person in the house and gets his own spinoff.
Hahahaha this guy was thinking he would only get ten views... next minute
56 is a small sample. If you could take the number of cases in every city it would be more accurate. Or it should be....
Very interesting.
Why is it that the numbers are distributed in this way?
Is there’s a short answer to that question?
Branford’s law requires a data set of numbers that span large orders of magnitude. For example Benford’s law will not work for the height of males in metres in a given population because almost all heights will begin with the number 1 and only a tiny number will begin with 2 and there will be no heights beginning with a 3 or higher. There may be dwarfs that are shorter than 1m - not sure.
The point is that you need to specify what parameter and range of measurement is applicable to the data set before you can ascertain whether Benfords law applies or is a useful analytical tool to deploy
What actual numbers or quantities did you choose?
Pr(d) = log(1+1/d)
Where Pr(d) is the probability of a particular leading digit, d, appearing in the data set.
If you substitute d=1 you get Pr(1) = log(2) = 0.30103 or 30.1%
Pr(2) = log(3/2) = 0.1761 or 17.61%
and so on..
Well we are actually detecting an anomaly that requires investigation of the underlying data. To say that it will detect a lie is in itself a falsehood.
Excellent presentation and choice of data!
Great easy to understand video. And no surprise that different states are reporting cases in different ways, not to mention the inaccuracy of the tests. One common mistake is to assume the underlying data for each data set member is incorrect when in fact, the results here only show the randomness or lack thereof for the data set as presented. This example illustrates that quite well.
Mr. Bonavito, for data to look suspicious, what ratio between observed and expected should I use as threshold? For instance, if +/- 25% is the fraud threshold, and the "1"s are outside of +/- 25% of the expected 0.3 (higher than 0.0375 or lower than 0.0225), then I would view the data as suspicious.
You had a keen observation.
Mr Bonavito used "eyeball" analysis.
The histogram did not look Benford was his conclusion.
You want a more quantitative analysis.
There are many, many ways to measure Benford quantitatively. The most common is the Chi Square test.
You asked for a ratio that deviates from the expected value. You are on the right track because it is an excellent test but unfortunately there is no simple ratio. You would have to calculate the Z score for all 9 digits.
To be 95% confident that Mr.Bonavito's example DOES NOT follow Benford the Z score should be greater than 2.56(because 9 digits are being tested the Bonferoni adjustment comes in to play).
I calculated the Z score for 3 digits and not any was over 2.56.
An example to calculate the Z score for leading digit 1 is:
(.41-.30)/((.30x.70)/56)^.5= 1.8
.41 is the actual percentage
.30 is the expected percentage
.70 is 1 minus the expected percentage
56 is the sample size
^.5 is the square root
All 9 digits can be tested that way using each digits own specific actual and expected percentages.
Distribution is the key . Like chicago . The way the make up is heavy blue districs go all at once it will spike numbers am I right ?
So the assumption is that because the distribution does not match the expected outcome, there is fraud likely involved.
But with Benford's law being a statistical instrument, it is tied to statistical annomylies that increase massivly for small sample sizes (anything below 1000 data points id argue is a small sample size). So at what point do you say "ok this is just not enough data points for me to make a judgement call on it" instead of "here is likely fraud in play" ?