Lies, Damned Lies, and Statistics

by Gerry Allan | Feb 15, 2023 | Forecasting, Managing | 0 comments

“Do not trust any statistics you did not fake yourself.”
— Winston Churchill

“Facts are stubborn, but statistics are more pliable.”
— Mark Twain

“It’s easy to lie with statistics. It’s hard to tell the truth without statistics.”
— Andrejs Dunkels

“Definition of Statistics: The science of producing unreliable facts from reliable figures.”
— Evan Esar

“In ancient times they had no statistics so they had to fall back on lies.”
— Stephen Leacock

“Everything that can be counted does not necessarily count; everything that counts cannot necessarily be counted.”
— Albert Einstein

“Do not put faith in what statistics say until you have carefully considered what they do not say.”
— William W. Watt

“There are many people with an agenda out there, and they will ruthlessly torture statistics until they get the answer they want out of them.”
— Malcolm Kendrick

“Skepticism is always a good first response.”
― Charles Wheelan, Naked Statistics

“Statistics: The only science that enables different experts using the same figures to draw different conclusions.”
— Evan Esar

“The essential feature of statistics is a prudent and systematic ignoring of details.”
— Erwin Schrodinger

I read last week about the amazing half-million-plus jobs growth in January 2023. What great news! Or perhaps not. Maybe it is little more than an opinion. Or possibly a lie, or even a damned lie, or worst of all – a statistic. What is the difference between these, if any? And why does it matter to us normal folk?

It was either the famous writer Mark Twain or the 19th-century British Prime Minister Benjamin Disraeli who said: “There are three kinds of lies: lies, damned lies, and statistics.” I love this quote, no matter who said it (if anyone), but the meaning is a bit obscure in practice.

The most common meaning seems to be: “lies” or “fibs” are generally harmless untruths; “damned lies” are dangerous untruths with malicious intent; “statistics” are worse yet because they may be true, untrue, or partially true, and it can be very hard or even impossible to determine which.

The last meaning is often applicable to statistics from almost any source.

The main problem with statistics is that people tend to believe them

As I have no doubt mentioned before, statistics like so many other things is just a mathematical tool. It can be very useful and reliable, or it can be very harmful and untrustworthy. Its nature depends on who generated the statistic and who might be its intended user or recipient.

In both cases, the problem is not the tool, but is either the creator or the recipient. That is, people are the problem. A hammer is a tool that can be useful or harmful depending on who is using it and who may get hit by it.

Most sensible folk would steer well clear of an Attila-the-Hun-type guy swinging a nasty big hammer or whatever. The potential danger of getting hit, accidently or deliberately, would be clear.

But what about a faceless bureaucrat or government agency that we pay for, releasing some monthly statistics on unemployment and hiring? Who actually reads and pays attention to these outputs? Very few people. Thankfully, these few include a very few who understand the numbers and can provide a useful translation – independently.

Who among those that take the statistics uncritically at face value might be damaged if the stats are misleading or worse? Well, many businesspeople who rely on such numbers for guidance on hiring and investment. They can make some very damaging decisions if these numbers are unreliable (aka wrong).

A specific example that triggered this post

Molly Smith via Bloomberg on February 3, 2023 on the latest greatest jobs report: “‘Too Good to Be True’ Jobs Report Draws Skeptics on Data Quirks”:

> Bloomberg Economics said payrolls surged on seasonal factors
> Others played that down, noting figures looked ‘pretty clean’”

“Economists are scratching their heads as to whether the US labor market is truly as strong as the latest employment report indicates or if wonky adjustments are getting in the way.”

“Employers added 517,000 jobs in January — nearly double the prior month’s advance and above all estimates in a Bloomberg survey. The unemployment rate also unexpectedly retreated to 3.4%, the lowest since 1969, according to Labor Department data released Friday.”

“Forecasters can at least agree on this much: the jobs market has shown few signs of falling victim to the Federal Reserve’s most aggressive interest-rate hiking campaign in a generation.”

“But the employment report was complicated by the government’s annual benchmarking process, as well as an update of seasonal adjustment factors and population controls that economists indicated before the release could make the data difficult to interpret.”

“For the establishment survey, the government’s updated seasonal factors may have impacted the headline payrolls figure. On an unadjusted basis, payrolls actually fell by 2.5 million last month [emphasis added].”

“Others aren’t so convinced that the revisions played much of a role. The 2.5 million decline in unadjusted payrolls was the smallest for January since 1995 — significant for a month that’s typically weak due to laying off holiday workers.”

“’We can’t completely dismiss all of these data. We can’t blame it on the seasonals,’ said Jennifer Lee, senior economist at BMO Capital Markets. ‘Breaking it down by industry, it’s pretty safe to say there’s wall-to-wall strength.’”

So, is this really good news or really bad news?

It surely matters if you are an executive or manager trying to figure out whether to lay off people or to start hiring again. What would you do in this case? How can you tell which story to believe?

Big stakes can be involved if you make the wrong move.

Note that nothing here is being hidden. The key summary points are readily available. Seasonal adjustments are standard practice in huge numbers of data sets. So, which to believe – the seasonally adjusted statistics showing a jobs gain of 517,000, or the unadjusted statistics showing an employment drop of 2.5 million?

The sad fact is that both statistics are “true”. The problem, not readily apparent, is that each one measures a very different thing.

There is also the matter of the official statistic possibly being aggressively-adjusted for political purposes, assuming that such farfetched happenings ever happen.

What’s really going on with the jobs data?

As an old analytics guy, I see immediately (or think I do) what is going on here and what an executive decisionmaker might be best advised to do. The problem is that not many folks out there are old analytics guys. You have probably noticed this. Most normal people pretty much have to take such official statistics at face value as a result. My sense about what is actually going on:

Big jobs gain: The 517-thousand jobs gain in January, post-adjustments and likely political tweaking, is at best reflecting some sort of longer term average jobs growth behavior. That’s what seasonal adjustments typically do. This view is helpful if you are comparing year-to-year periods. The adjustments help remove the “noise” that is nearly always present in most short-term data sets.
Big jobs loss: The 2.5 million jobs loss, from unadjusted raw data, reflects what is going on right now. It’s what the current set of source data from reliable (we hope) sources are reporting. Unless these sources have suddenly become unreliable for some unknown or unreported reasons, the raw data set is especially important in these times of great, rapid change. What one might logically do at this point is to wait for confirmation from the February raw data. Or maybe not – see below.

For any kind of serious business decision involving major investments or major layoffs (or hiring), you probably want to look first at the behavior of unadjusted and un-fiddled-with data. Not that the likely data-fiddlers might have anything that requires serious fiddling at this moment, but focusing on raw data initially removes any concerns about possible fiddling.

One might also consider the relatively large layoffs reported by big companies in recent months. These typically are only a few percentage points of total employment, but they may be just the beginning.

To me, the current situation would strongly counsel caution.

Many data sets support multiple honest interpretations

Despite the widespread, well-justified skepticism concerning both raw (unadjusted) data and related statistics from almost any “official” source, the reality is that the meaning is often far from clear. You can get this situation even from looking at a single dataset and its statistics.

Analysts and normal people have widely different viewpoints on pretty much everything. Such differences can bias the way in which a dataset-statistics statement might be understood and interpreted. You don’t even need to have different agendas involved, just different viewpoints and lenses based on different knowledge, experiences, opinions, and many similar personal factors.

In fact, it may well be rare for any general agreement on meaning to occur.

It gets much worse: What is the real rate of consumer price inflation?

While it is quite clear that prices for almost everything are going up, and have been going up for at least several years, there is a wide range of opinion on the real consumer price index (CPI) inflation rate. The official government CPI-U inflation rate today is just above 6% on an annual, year-over-year basis. The “-U” suffix stands for urban population prices, as published each month by the U.S. Bureau of Labor Statistics (BLS). A recent BLS chart with major categories is shown below.

BLS Consumer Price Index across major categories.

On the surface, 6% a year seems high but tolerable. A basket of goods that cost us $100 in December 2021, costs $106 in December 2022. If this rate of increase, aka price inflation, continued at this rate for 10 years, we would be paying $179 in December 2031 for that same basket of goods. Not bad if our income growth at least roughly matches this price growth rate (see here for more detail).

John Williams’ ShadowStats CPI-U charts for the past 20 years shows official CPI-U growth rates of 2% to 4% over most of this period. They also show alternative rates based on calculation methods used prior to 1980, and pre-1990, which are much higher – roughly double the official rates. Which rates reflect reality?

http://www.shadowstats.com/alternate_data/inflation-charts/

The answer here is vitally important since it drives a great deal of government and business policy and actions.

Timothy Lee in his Full Stack Economics Substack blog from November 2021 argues that Williams hugely overstates actual price inflation: “No, the real inflation rate isn’t 15 percent”:

“Last week, the Bureau of Labor Statistics announced that the official inflation rate had soared to 6.2 percent in October, the highest level in decades. But some … analysts insist that the true inflation rate is much higher. Last Wednesday, for example, self-help guru Jordan Peterson tweeted a chart that purports to show the year-over-year inflation rate is almost 15 percent, not 6 percent.”

“The chart comes from a website called Shadow Government Statistics. Its premise is that the Bureau of Labor Statistics made a series of methodological changes in the 1980s and 1990s that have systematically understated the true rate of inflation. According to ShadowStats, if you calculate the inflation rate using old methodology from the 1980s, the true inflation rate is 6 to 8 percentage points higher than the official statistics indicate—and has been for decades.”

“But the creator of the chart, economist and ShadowStats founder John Williams, has admitted that he doesn’t actually re-compute the inflation rate using earlier methodology. He does something much cruder: he starts with the official inflation figure and adds a fudge factor that represents his estimate of how much the official consumer price index (CPI) understates the true inflation rate.”

“The problem is that Williams’s adjustment is way larger than it should be. The BLS has changed its methodology over the years, but all of those changes put together have probably changed the measured annual inflation rate by a fraction of a percentage point—not the 6 to 8 percentage points Williams claims.”

Price inflation looks pretty real and serious if you buy things like food

Sorry for yet another chart: This one from CNBC shows annual price increases for many categories of consumer goods. The bottom line here shows that the official 6.3% price inflation excludes minor purchases such as food and energy. Who buys much of that stuff anyway?

The raw data that BLS uses is probably real. Especially since the chart shows huge price increases in categories where normal folks spend most of their money. The exclusion of “food and energy” from the CPI statistics is based on these components being highly variable, and thus are considered by the BLS as misleading “noise” for their purposes.

Extract from CNBC chart based on BLS figures for October 2022.

The BLS figures clearly aren’t “lies”, or even “damned lies”. They are “statistics” developed for government purposes, which are neither hidden nor unreasonable – in context. But the BLS figures are, for the purposes of us normal folks, almost meaningless. Try building your spending budget for the coming year on BLS official price inflation statistics (e.g., 6.3%), and then on category detail statistics. Which one would you regard as “real”?

Official CPI statistics drive all manner of government spending

One possible reason why the BLS ignores minor purchases like food and energy is that they desperately need to understate true inflation. Official rates determine Social Security increases and government pay increases among so much else. They simply can’t afford increases based on what we actually are spending.

So they aren’t exactly lying. They are simply “justifying” an affordable rate based on government funding realities. The problem here is that they don’t make much if any effort to explain what they are doing. Imagine the uproar if they said that inflation is actually 15%+ annually right now but we can afford to deal with only 6% or so. Have a nice day.

The problem for us is that so many people simply believe whatever the government tells them. The problem is “us”, as noted in a recent post.

You undoubtedly have heard this old saying:

“You can fool all the people some of the time, and some of the people all of the time, but you cannot fool all the people all of the time.” — routinely misattributed to President Abraham Lincoln, but probably originated – in French – by Jacques Abbadie in 1684.

In practice, this observation should really read “… you can fool enough of the people some of the time …”. You don’t need to fool all of the people all, or even any, of the time. People have an inborn tendency to want to believe authority figures no matter what common sense might otherwise say. People also have a tendency to avoid the often hard work of finding out what is actually going on. Some sizable percentage simply does not want to know, or does not really care.

These true believers are generally willing to believe official statements and statistics.

Going along to get along is the general rule in life

This may seem a tad cynical but there is research by experts such as our good buddy Mattias Mass-Formation-Psychosis/Hypnosis Desmet to support such conclusions. I have addressed this topic in several previous posts (see for example here, here, and here) so it is not worth repeating, except for these relevant findings:

“… we identified three groups that form when a mass rises: the masses themselves, who truly go along with the story and are “hypnotized” (usually about 30 percent); a group that is not hypnotized but chooses to not go against the grain (usually about 40 to 60 percent); a group that is not hypnotized and actively resists the masses (ranging from 10 to 30 percent).”

The 30% true believers may often be entirely adequate, but if not, some portion of the go-along-to-get-along crowd of 40% to 60% will certainly jump on board to provide the “enough” for current purposes.

Recession now or not? Let’s see what the statistics say …

First, we need the “official” definition of a “recession”. From Wikipedia:

“In a 1974 article by The New York Times, Commissioner of the Bureau of Labor Statistics Julius Shiskin suggested that a rough translation of the bureau’s qualitative definition of a recession into a quantitative one that almost anyone can use might run like this:”

“In terms of duration – Declines in real gross national product (GNP) for two consecutive quarters; a decline in industrial production over a six-month period.”

“In terms of depth – A 1.5% decline in real GNP; a 15% decline in non-agricultural employment; a two-point rise in unemployment to a level of at least 6%.”

“In terms of diffusion – A decline in non-agricultural employment in more than 75% of industries, as measured over six-month spans, for six months or longer.”

“Over the years, some commentators dropped most of Shiskin’s “recession-spotting” criteria for the simplistic rule-of-thumb of a decline in real GNP for two consecutive quarters.”

“The Bureau of Economic Analysis, an independent federal agency that provides official macroeconomic and industry statistics, says ‘the often-cited identification of a recession with two consecutive quarters of negative GDP growth is not an official designation’ and that instead, ‘The designation of a recession is the province of a committee of experts at the National Bureau of Economic Research [emphasis added]’.”

Got that? The official definition of a recession is whatever we at the NBER say it is. That’s not helpful, especially to folks who have to make some serious and potentially costly decisions regarding current economic conditions. Statistics from NBER are therefore useless or worse for real business purposes.

Recession indicators from real analysts

What this fudging of official statistical definitions does is to force decisionmakers (or at least I hope so) to ignore the official definitions and to look instead at the underlying data.

Real analysts create their own statistics if no reliable ones are available, but they often step back to use statistics that are less likely to be fudged or fiddled. One of my favorite analysts of this type is Mish Shedlock, who has a blog called MishTalk.com. Here is a quick look at what he wrote in mid-January of this year: “December Was Another Retail Sales Disaster, Even Worse With Negative Revisions”:

This looks pretty down to me, even if the experts at NBER missed these subtle clues. The chart below shows that October 2022 appears to be the actual beginning of a downturn.

There are many other valuable statistics for what’s really going on

Some of these are rather obscure. Here is an especially good one: the relationship between the number of people employed full-time and the number employed part-time. Both groups are “employed” but they have entirely different economic situations and impacts on the economy.

Ryan McMaken via Mises.org and Infowars presents the story here rather nicely: “Another Recession Sign: Part-Time Work Is Growing Faster than Full-Time Work”:

“Part-Time and Full-Time Employment Have Inverted, and That Means Recession. In recent months, employment growth has increasingly been driven by part-time rather than full-time employment. Since September, in fact, month-to-month employment growth in full-time jobs has been negative, while growth in part-time jobs has been positive.”

“For example, for much of 2018, year-over-year growth in full-time jobs numbered in the millions, while part-time employment actually fell. Sometimes this situation reverses. Indeed, a switch from full-time-driven employment to part-time-driven employment usually indicates that a recession is coming. We saw it happen in 1981, 1990, 2001, 2008, 2020. Now it’s happened again in 2023.

How the full-time to part-time employment changes from growth in full-time dominating through mid-2022, to part-time dominating from mid-2022 through today.

The fact that few analysts, for public purposes at least, follow such statistics does not diminish their fundamental value. Does the above story reflect a 517-thousand official jobs gain in January, or a 2.5-million unadjusted-data jobs loss that month?

Statistics in general have a purpose and essential context

Relatively few if any statistics are generated for fun. They are typically hard to create and interpret reliably. For this reason, statistics have a specific purpose and are valid only within a specific context. To use any of these with confidence, you have to know both their intended purpose and the conditions under which they have sufficient validity.

Statistics generated by government agencies are typically driven by politics and interest groups of some sort. Using these without a clear understanding of this powerful non-data-based context is likely to mislead and to cause harm.

Better instead to set them aside where they can neither mislead nor cause harm.

Bottom line:

The amazing half-million-plus jobs growth in January 2023, a statistic, is certainly real, but is not applicable to our current reality. It measures something quite different – a noise-suppressed longer-term view. We have to deal with the here-and-now in business. The real-real news is that the raw data on jobs shows a 2.5 million jobs loss last month. Furthermore, a number of other credible indicators supports the outlook for a serious recession being underway despite official denials and happy-talk. If you have any significant risks in taking major actions in the coming months, the official-non-lie-but-wrong-statistic story is useless, or worse.