Truth About NICS Checks

Interesting! Thanks for posting that.

You can clearly see the tremendous anomaly caused by Obama's election,

Click to expand...

Huh? I see no data from 2008 and before. It should start at 2004 at the very least if you want to make this point.

To account for general population growth it should be NICS checks per capita instead of raw checks, this could negate or reverse the "trend".

What is your criteria for "significantly different from the trend" when most (all but six by my count) of the data falls within +/- 3 sigma?

I agree that the seasonal pattern does seem obvious.

I guess the graph is okay as a general overview, but it leaves out my last 25 purchases-- no NICS here with a carry permit. Other states have that too, so multiply me by a whole bunch of people, many of whom are "into guns" and thus might buy more than one. Considering that permit numbers are going up too, the graph is going to be skewed downward a bit. Still nice to see a general upward trend overall, though.

Interesting post and data. One quibble:

denton said:

You can clearly see the tremendous anomaly caused by Obama's election...

Click to expand...

Since Obama was first elected in 2008, no, you can't. What you can see is the tremendous anomaly caused by the Sandy Hook massacre and (more importantly) the immediately-subsequent gun control discussion by the administration and senate. If Obama had remained as relatively-agnostic towards guns in his second term as he had in his first, the craziness of the shortages and panics and whatnot would have been trivial compared to what we got. (Democrats probably would have held onto the senate in 2014, too, but that's off-track...)

AZAndy said:

I guess the graph is okay as a general overview, but it leaves out my last 25 purchases-- no NICS here with a carry permit. Other states have that too, so multiply me by a whole bunch of people, many of whom are "into guns" and thus might buy more than one. Considering that permit numbers are going up too, the graph is going to be skewed downward a bit. Still nice to see a general upward trend overall, though.

Click to expand...

That's a very good point. I hadn't thought about the fact that because both my wife and I have CCW permits, none of the gun purchases either of us has made in probably 20 years show up on that chart.
And you're right - if permit numbers are climbing like we hear, "the graph is going to be skewed downward a bit." Probably not much though, because I suspect a good sized number of new CCW permit holders bought their one and only gun, the same one they used to qualify, before they applied for their CCW permits.
However, Idaho and several other states have gone Constitutional Carry now. So that might skew the graph back upwards a bit. I'm getting confused!
It's still interesting. And like you, I think "it's still nice to see a general upward trend overall."

If you're seriously interested in normalizing the chart for population

Click to expand...

I'm not interested in charting anything, but your claim of a "tremendous anomaly caused by Obama's election" visible in the chart is totally bogus without a significant amount of data pre-2008 to back it up.

Data from https://www.fbi.gov/file-repository/nics_firearm_checks_-_month_year.pdf/view . Except for the partial data from Nov-Dec 1998, that's all there is.

The best fitting trend of the simple ones available in a spreadsheet was an order 2 polynomial, suggesting that sales are not just growing, but that the growth is accelerating. ...but trying to forecast from data this noisy and apparently event driven is, shall we say, fraught with peril.

.308 Norma said:

That's a very good point. I hadn't thought about the fact that because both my wife and I have CCW permits, none of the gun purchases either of us has made in probably 20 years show up on that chart.
And you're right - if permit numbers are climbing like we hear, "the graph is going to be skewed downward a bit." Probably not much though, because I suspect a good sized number of new CCW permit holders bought their one and only gun, the same one they used to qualify, before they applied for their CCW permits.
However, Idaho and several other states have gone Constitutional Carry now. So that might skew the graph back upwards a bit. I'm getting confused!
It's still interesting. And like you, I think "it's still nice to see a general upward trend overall."

Click to expand...

Yes. But then there's people like me who got their CHL to make buying guns easier. Not just NICS, in NC you have to get a purchase permit from the local sheriff for every handgun you buy (retail or private seller). But with a CHL that isn't needed. Since getting my CHL I've bought 7 or 8 firearms?

Point being, I agree with you that it's hard to tell just from NICS.

So exactly why is that so?

Click to expand...

You've made a claim of a "large Obama effect" and presented absolutely zero data to back it up. I despise made up "statistics" even if I support the cause or the idea behind it.

The best fitting trend of the simple ones available in a spreadsheet was an order 2 polynomial,

Click to expand...

Polynomial models are almost never good for anything other than interpolation to estimate what data you don't have (or couldn't measure) might have been, but generally splines or sync interpolation is "better". Polynomial models are terrible for "extrapolation" beyond the observed data because they ultimately always diverge to infinity or zero. Its classic freshman stats class to fit polynomial models to human world population data and extrapolate to find the "day of reckoning" when the earths population equals zero, or the "day of doom" when the mass of humanity would equal the mass of the earth!

The "best" trend fits are Box-Jenkins ARMA models stopping at the lowest order that has "white noise" residuals. The anti's are big on using these models on data that doesn't support their argument by increasing the orders until the model shows what they want which is incorrect application of the method and a classic example of "how to lie with statistics".

I'll agree with the others - this chart proves exactly NOTHING to me. That statement of "little red numbers indicate a deviation from trend," coupled with little red numbers appearing on about HALF of the data set largely suggest the "trend" doesn't hold true for the data set. I'll also agree there's no correlation here between Obama taking office.

What I see in both the original and subsequent charts, is an increase in volume and an increase in volatility.

It's disappointing to consider how many private sales and pre-approved CCHL sales are not incorporated, and of course, these numbers reflect ALL transfers, not new firearms sales, as it further removes these charts from correlation with realistic trends in new firearms ownership - and of course, not all NICS checks signify new firearms owners, only signify FFL transfers to a non-CCHL holder.

Of course, if you consider in that timeline the open market evolution of the AR-15, the sunset of the AWB, the increase in media sensationalization of foreign and domestic terrorism over the time period, the evolution of online firearms sales, a DRAMATIC increase in concealed carry permit issuance and marked changes in state laws allowing concealed carry, etc, it becomes very easy to pick any of a number of reasons for the transfer volume increase. But none of these charts show a strong correlation with an election preparatory OR election reactionary responses in sales volume. If folks were buying as a response to Obama's pending or confirmed election, there would be a greater correlation to high volume leading up to and immediately following those election years, then a stabilization mid-term, with a logically lesser response in kind prior to the second election - with a tail off following Trump's election. But that trend doesn't appear.

What interests me, and what I might say is indicative of a LACK of correlation between sales and election cycles - there's a high trend in early 2016, however, sales fall off in May/June of 2016, as we were approaching closer to the Hillary/Trump election. Personally, I would have expected a much greater grass roots response via increased gun sales in preparation for a potential Hillary term than that of an Obama term, so that tailing off right before the election, close as it was, especially with her leading in the polls prior to the election, just doesn't support a correlation between sales and election fears.

I'd be more interested in a comparison of birth rates ~18yrs prior to each year - we're growing very quickly as a country, and we're seeing a great disparity in birth rate growth between the last few generations - if the baby boomers are going to take down social security simply because there were way too many of them compared to the generation which went before, it's easy to understand how birth rates for a given decade might explain away a trend like this.

My question was, why is there no acceptable basis for comparison without 2008 data? That question goes to your original objection, and stands unanswered.

Click to expand...

Only because you've not answered it. You've claimed your chart shows an "Obama effect" with no data pre-Obama to back it up and show there was any effect whatsoever. Plain and simple. If you don't understand that I give up.

Your claim of extraordinary variation of single data points that made up the data set to calculate the trend is suspect, and if it is somehow "real" you've nothing but speculation as to cause.

In the full data set plot shown by DrDeFab he could at least wave his hands and claim a near constant rate of checks from 1999 to 2004, with a small upslope in checks from 2004 to late 2008, and a larger upslope from 2008 (with larger fluxuations) to present. There is not enough data post Nov 2016 to claim the slope has increased or decreased. I can wave my hands and say this was "caused" by the AWB
expiring in 2004 and Obama's election in 2008 (or was it the market crash? or Hurricane Ike?) but without real statical modeling and analysis its all BS and on no more solid ground than the claims that criminals get their guns from gun shows and therefor we need universal background checks.

As you say, looking at the full data set, an exponential underlying trend is obvious

Click to expand...

A polynomial of order 2 is a long way from being exponential.

denton said:

That statement of "little red numbers indicate a deviation from trend," coupled with little red numbers appearing on about HALF of the data set largely suggest the "trend" doesn't hold true for the data set.

Are you seriously proposing that there is no long-term underlying trend in the data?

Click to expand...

No - I'm saying exactly what I said - there is no reason to suggest the trend line is accurate to the data set when nearly half of the data is listed as excursions.

If half of your data includes excursions, then the excursions are not excursions, they are a significant contributor to the data trend.

wally said:

A polynomial of order 2 is a long way from being exponential

Click to expand...

There's a large contingent of the country which confuses an exponential trend with a power factor trend.

denton said:

A polynomial of order 2 is a long way from being exponential.

Excuse me? A second order polynomial is defined as one that contains x^2 terms. 2 is not an exponent in this case?

Click to expand...

No - a 2nd order power factor trend is not an exponential trend.

A second order polynomial: y = x^2...

An exponential trend: y = A^x

Very, VERY big difference. High school algebra isn't typically most a favorite for most people, so it's common for folks to mistake a trend which contains a constant as an exponent on the variable - a power factor - with a true exponential trend in which the variable is the exponential term.

Containing an "exponent" does not make a trend exponential. It makes it a power factor, or N order polynomial.

An exponential function has the variable AS an exponent. An n order power function has the variable to the power of n. Y = A*e^X is indeed an exponential function, because the variable, X, is in the exponent. Y = X^2, a second order polynomial POWER FUNCTION, is NOT an exponential function, as the variable is NOT a component of the exponent.

denton said:

But I don't think you can properly exclude fixed exponents from being exponential.

Click to expand...

Yes, indeed you can. Functions including only fixed exponents - aka constant exponents - are indeed NOT exponential functions. That's the very definition of exponential functions as compared to that of power functions.

Go back to high school.

denton said:

One position a person might take is that there is no trend. In that case, you just lump all the data and take an average.

The opposing position is that you can get a better estimate of the situation by extracting the trend that is in the data, and looking at scatter around that trend.

The first step is to decide which of those propositions you like better.

If you think using a trend is better, then the usual choice is for the trend that predicts the data best. You can pick an exponential, or you can pick something linear, or logarithmic if you like. Over a short span, linear usually works well. The centerline shown is the linear choice that produces the least error in the centerline.

What is there about that that you find objectionable?

Click to expand...

This is largely gibberish - the data presented is a time series, so an average over the entirety of the data set as a comparison for time dependent trend makes no sense. Sure, if you wanted to say over the last 7yrs, there have been an average of XXX transfers per year, that would make sense, but for a time dependent trend, that doesn't make sense.

1 ,1, 2, 2, 1, 2, 1, 3*, 2, 1**, 2, 3, 3 4*, 2**, 3, 4, 4, 3, 4, 5, 4, 5, 3**... Consider this trend of time values. I starred those which don't fit with the trend prior to it (this is a simplification of an ACTUAL algorithm some data loggers use to compress data). In this series, there's an upward trend. However, I have exclusions noted any time the current value exceeds or undershoots the previous value by too great of span. So even though these local excursions very well may fit within the noise of the series, which is an upward trend, any time it jumps too quickly or falls too quickly from one value to the next, an excursion is recorded. Is that apt? It's an upward trend, with some noise. Should those starred or double starred values be excluded? It's clear a 1 at the end of the series or a 5 at the beginning would be an excursion, but a 3 anywhere shouldn't be an excursion, nor a 2 at the early phase, or a 4 near the end, yet they're marked as such... So what are the red numbers really indicating? Excursions from the trend, or real contributors to the overall trend?