Does QUICKNESS do anything? An empirical study...

[deadfaulkner]

Over the past six months, cheapcatch, slimstud7, and I have been operating a TSB-III league called the "Tecmo Legacy League" - a name we came up with because it's a progression-style league set whose inaugural season is set in 1976. All 28 teams (and therefore all 896 players) are MAN controlled, and this has meant a LOT of research, which has been quite a bit of fun actually, and a lot of time trying out methods to turn historical information in to TSB-III style attributes.

We've been very thorough about it all, and are actually quite proud of the work we've done so far. We know what a 44 PC quarterback looks like, a 63 MAX SPEED running back looks like, what a 44 REC tight end looks like, and so on. But one thing still gives me fits - what on earth is the difference between an 81 QUICKNESS linebacker and a 38 QUICKNESS one? I really have no idea, and it makes me feel a little silly, having done all this research and not knowing this one.

It's not solely down to the fact that I'm not the sharpest knife in the drawer that I don't know what this attribute does - even the hurricanes dude who is, in part, responsible for those crazy-hacked NCAA ROMs says that QUICKNESS is "undefined." And, given that I'm nothing close to the hex-editing genius that those guys are, I'm not going to get the answer to this question from the game's hex code. So then, the only other way to attempt to isolate what QUICKNESS might do is empirically.

The first thing I set out to do was to make two teams that were as equal as possible to establish a control test. On the surface, this seems pretty easy. I took the two teams we're not using, the Jags and Panthers (which I renamed simply to X and Y) and gave them generic rosters. I set each player at each position to TSB-III median-level attributes (for example, the median PASS SPEED for a quarterback in TSB-III is 44, so I set the quarterbacks on X and Y to 44 PS), and scheduled the two teams to play each other in each week of the season, alternating home games (though I'm fairly certain home-away status doesn't do anything). Since auto-skipping wouldn't do to study the attributes (given that SKP game stats are figured from a single hex number, completely separate from attributes), I set the emulator to high-speed, and set both teams to COM.

I said that setting up two equal teams was simple on the surface because there are still two variables I couldn't control for. One, obviously, is physical condition, but I figured those changes are random (or random enough), and I'd planned to sim through enough seasons that it would normalize over time. The other, which would not normalize over time, was the suggestion of a "hidden" QB attribute, which has been mentioned in the past by, I believe, drunken_honkey (though, for the life of me, I can't find the thread now), which basically determines how smart the AI abs will be in choosing a receiver. I've never bothered with this, since our league is an all-MAN league, and AI quarterback decision-making never matters. But I didn't know how to go about finding this in the hex, so I did go into these tests with some concern that this hidden attribute would skew the results. Turns out, though, that one of two things happened: either the hidden att for the QB on each team was the same or similar, or this att simply doesn't do much, as the two teams produced similar enough results over time.

For the control, I set the games to fifteen-minute quarters, and had no set idea for how many seasons I'd sim - at least a few, I decided, to get a sufficiently large sample, and from there, as many more as it took to normalize the results - i.e. until the total stats for each team were even, or even within an acceptable range. That happened after seven seasons.

My initial thought was that QUICKNESS has something to do with passing defense, and that, since it does not appear to do anything visible in the game, it's a calc stat, in the same way that INTERCEPTIONS or PASS ACCURACY are. I had read somewhere along the line that, for example, you could create a "shutdown"-style corner by giving him high QKN and lower INT. I'd also read elsewhere that raising QKN increases the likelihood that a player will leap and make a deflection. So I was first going to be looking at passing statistics. Here were the overall passing totals from the control test:

Comp % YPA TD% INT%

45.70 7.52 6.53 5.10

I then decided to examine the effect of increasing a known attribute - INTERCEPTIONS. It seems pretty clear that raising INTERCEPTIONS leads to an increase in, well, interceptions, and it also generally held that it, along with PASS ACCURACY and RECEPTIONS, also is calculated in determining whether a covered pass will be completed. So I raised the INTERCEPTIONS attributes for all defenders on one of the two teams to 100 and then ran three seasons of sim (with a view to simming more, should, for some reason, the results be inconclusive). Here are the passing results for that test for the team of that faced the 100-INT defenders:

comp% YPA td% int%

33.63 6.34 4.44 26.41

Yep, no surprises there, and no need for more sims: interception percentage skyrocketed by over 5x, and everything else dropped.

Well, maybe there is one surprise. If you subtract the number of interceptions from both tests, and just look at the results of non-interception passing plays, you get some interesting results. First, we take the interceptions out of the equation for the control:

Comp % YPA TD%

48.16 7.93 6.88

Then for the 100 INTERCEPTIONS test, with interceptions subtracted:

comp% YPA td%

45.70 8.62 6.03

The surprise here for me was how little the difference between the two results was - especially in completion percentage. Of course, I was simming at high speed, and couldn't actually count the number of times the quarterback threw into coverage, but from plenty of anecdotal experience with the CPU, I know with certainty that they throw into coverage a LOT. If INTERCEPTIONS is one-half of the equation for a completion, as it is for an interception, you'd expect that the CPU quarterback would produce incompletions at a rate similarly askew from the control test as he did with interceptions. That, however, did not happen: the difference in completions was small, which leads me to think that perhaps the completion/incompletion calculator lies somewhere else: perhaps QUICKNESS?

So, finally, I did a test for QUICKNESS. I set the defenders back to TSB-III-median INTERCEPTIONS, and raised the QUICKNESS for each of them to 100, and ran the test, once again covering three seasons. Here are the passing results for the team opposing the raised-quickness guys:

Comp% YPA TD% INT%

45.34 7.98 7.90 4.85

Yep, that's right - these numbers are fairly identical to those produced in the control test. And it's certainly a large enough sample: we're talking about 2393 passing plays here, more than enough to produce statistical variation if QUICKNESS had anything to do with passing defense. In fact, each of the major passing categories (completion percentage, yards per attempt, touchdown percentage, and interception percentage) fell within the standard deviation, meaning that there was not a single incident of a statistically significant deviation from the control test seasons.

We can conclude, then, that QUICKNESS has nothing to do with passing defense.

So, might it have to do with something else? How about rushing average? Maybe QUICKNESS makes defenders, well, quicker, and therefore able to get to the ball sooner and, over time, prevent longer runs. A nice theory, I guess, but a false one - the control rushing average is 5.47, and in the QUICKNESS test, the opposing offense averaged 5.57 yards - again, well within the standard deviation. Receiving average is identical, too (17.57 in the control, 17.59 in the test) though if it wasn't, I don't know how a raise in QUICKNESS could have explained it.

How about fumbles? I have heard the theory that high QUICKNESS is a calc attribute, much like INTERCEPTIONS, but for fumbles. We can test this if we add together pass completions and rushing attempts and subtract from that total the passing touchdowns and rushing touchdowns, giving us the total number of plays that, presumably, ended in the field of play (the CPU doesn't run out of bounds much). Then if we divide the total number of forced fumbles by the defense by that total, it will give us the fumble percentage.

For the control:

ffbl Plays Fb%

485 12521 3.87

For the test:

ffbl Plays Fb%

119 3394 3.51

Nope. Fumbles are relatively rare events, for sure, so you might argue that there isn't a large enough sample size to accurately judge. But I think it's a fair assumption that if QUICKNESS had anything to do with the probability of forcing a fumble, there would be at least enough evidence to suggest the need for more tests. But, once again, all the test seasons fell within the standard deviation derived from the control, and that is enough for me to conclude that QUICKNESS has nothing to do with fumbles.

So, we can draw a number of conclusions from this. For one, QUICKNESS is not a calc stat of any kind, because it would have shown up in the CPU vs. CPU statistics somewhere. But every kind of manipulation of the numbers from the QUICKNESS tests shows the results well within the standard deviation, without a single significant statistical variation.

There remains the possibility that QUICKNESS is a "physical" attribute, like MAX SPEED or RUSHING POWER, and that it does something visible in games. Until someone invents a TSB-III Scouting Combine hack, though, no empirical test that I can think of to test that. So we're left with anecdotal evidence, and as someone who's played thousands of games of TSB-III (and I'm pretty sure that if you're reading this you've played plenty of games yourself), I am quite confident in saying that there is no visible effect of QUICKNESS on a player's movement or performance.

So, I think we can answer the original question - does QUICKNESS do anything? No, it does not.

But one of the more interesting things that this study also found is that the effect of INTERCEPTIONS on whether or not a pass is completed is minimal - much less than I'd ever thought it to be, for sure.

So, boys and girls across TecmoLand - go ahead and throw at Rod Woodson with reckless abandon. Yep. Just go ahead. Do it.

[Knobbe]

It's an interesting study....sort of odd if they still didn't use Quickness. I suspect someone either has or will decode the game play itself....it will be interesting to see how this study will compare.

BTW, while you might have concluded that passes were completed at the same rate against high INT DBs, you seemingly forgot that that INTs do count as incomplete passes...in fact worse.

So, boys and girls across TecmoLand - go ahead and throw at Rod Woodson with reckless abandon. Yep. Just go ahead. Do it.

Yes...please continue to throw at Rod Woodson

[bruddog]

Can you record in the znes emulator? I can't remember.

That was how i determined quickness did nothing in the NES. I would record a game with one set of quickness values...say every defender set at 81 quickness..

Then i would go in and change the values to 13 and replay the movie using the updated rom. Since quickness did nothing the game didn't desync and become gibberish like it would if you changed an attribute that was used.

Playing devil's advocate as quickness may do nothing in TSB3 as well but here is one thought.

1. Quickness is only used if the player is man controlled. Seems unlikely but that's a thought.

************************************************************************************************************************************************************************************************

As far as the completions interceptions things I think you are confusing things but it would take me a long post to explain it. But 'll make one quick point.

It's very likely that even with COM passing a large number of the completions were thrown to an open man or thrown it so that the wr had to jump for it. My reason for this is as follows: On the NES rom if a 56pc throws to a 69rec wr with a 100int defender. If its a true coverage catch attempt where the WR doesnt dive or jump and the defender is in position, you would except 68% defelections, 0% catches, and 32% interceptions.

Since TSB3 likely works in a similar way you'd have to figure most if not all the completions the defender int was even used because he defedner wasn't close enough or it was a jump pass.

*********************************************************************************************************************************************************************************************

[DrunkenGamer]

The hidden rating for QBs is the scramble code. The last byte of the QBs ratings. It doesnt show up on the games ratings, but is used for COM QBs and Sim. Its written in the QBs pass coding. If high enough, the QB will take off and run. We too could never see or find any use of the Quickness rating.

[ Post made on a Tecmo Super Mobile Device!]

[TecSpectre]

I was researching quickness a short while back, and I found an article on this site posted a couple years back about all of the stats. After reading that we tested out what it stated and found that quickness does have an effect in TSB III...

Quickness is a speed factor that affects the speed at which a man controlled player will move toward a receiver after a ball is passed, when the computer takes over control of the player, while the ball is in the air.

[Eric]

Wow, that is good to know!

[Knobbe]

Can you record in the znes emulator? I can't remember.

That was how i determined quickness did nothing in the NES. I would record a game with one set of quickness values...say every defender set at 81 quickness..

Then i would go in and change the values to 13 and replay the movie using the updated rom. Since quickness did nothing the game didn't desync and become gibberish like it would if you changed an attribute that was used.

Playing devil's advocate as quickness may do nothing in TSB3 as well but here is one thought.

1. Quickness is only used if the player is man controlled. Seems unlikely but that's a thought.

************************************************************************************************************************************************************************************************

As far as the completions interceptions things I think you are confusing things but it would take me a long post to explain it. But 'll make one quick point.

It's very likely that even with COM passing a large number of the completions were thrown to an open man or thrown it so that the wr had to jump for it. My reason for this is as follows: On the NES rom if a 56pc throws to a 69rec wr with a 100int defender. If its a true coverage catch attempt where the WR doesnt dive or jump and the defender is in position, you would except 68% defelections, 0% catches, and 32% interceptions.

Since TSB3 likely works in a similar way you'd have to figure most if not all the completions the defender int was even used because he defedner wasn't close enough or it was a jump pass.

*********************************************************************************************************************************************************************************************

Yeah, basically record a game you play against the computer.

Play it back and verify that it syncs up for sure.

You could then change some sort of non-game play thing like player race or uniform color and retry again.

Then change the quickness attribute on all your players. If the playback syncs up then quickness doesn't do anything. If the results are unexpected, then it does.

[DrunkenGamer]

Or just find the attribute sliders in hex, make it 00, watch a game. Then change it tp FF, watch a game. See if anything is noticably different.

[DrunkenGamer]

By changing it to 00 and FF, I mean all ratings equal 00 or FF.

[pjfoster13]

I have had the original TSB since I was a little kid and still play the game on smartphone emulator, I've probably simulated 10,000 seasons of this game... I am not a programmer but I think I have the answer to this question. 

 

Quickness is not a USER-functioning attribute but rather an input into the sim itself when you simulate seasons on SKP v SKP. It controls the distribution of Sacks and Interception yards (and TD) under simulated conditions

 

Practical MAN v MAN and MAN v COM (or even COM v COM) gameplay is a completely different statistical experience than SKP v SKP. Let's say you're playing a MAN v COM with the 49ers (or whoever, pick your team), and let's say you prefer controlling the secondary guys like Lott or Waymer. Basically the only way your team is going to get a sack is if you blitz, and virtually every single time you blitz it's going to be Matt Millen (or whoever is your LILB) who gets the sack. This experience is typical no matter which team you play as, unless it happens to be a team with a dynamite RILB. The only other way to get a sack is if you have somebody like Bruce Smith or Lawrence Taylor who popcorns the offensive lineman.

 

HOWEVER, consider how the statisical outputs function when you sim every game on SKP v SKP. Sim any sample season with the 9ers, and you will notice certain trends. One conclusion we can make is that the "LILB gets the sack on all pass blitzes" effect completely disappears. The ILBs get a sack here or there but nothing much. Also, the game completely ignores the "popcorn effect" because NT Michael Carter has high HP but typically accumulates very low sack totals, even when playing teams with weak centers. The sim does not "match up" the HP of OLB v T, DE vs G, NT vs C in the same way that it does when you are actually playing

 

What I do notice, however, is that season sack totals correlate to Quickness (+MS). Charles Haley is a demigod on the sim, his quickness (+MS) is ridiculous and reflects in the season sack totals when you sim, but he's not always as good when you play. However, there are a few names I always noticed on season sim and wondered why- Broderick Thomas (TB), Ken Harvey (PHX), Jacob Green (SEA) are names who come to mind. They're all decent players but nobody who is ever going to popcorn unless under EXCELLENT v BAD conditions; definitely not the world-beaters they end up being the way they are on SKP sim. If you were to play COM v COM with either of those teams, you might play through 100 games before either of them actually gets a sack, but in SKP seasons they regularly get a sack per game. 

 

So in essence, I am inclined to believe that the sim engine goes thru play by play and assigns a certain number of "blitzes" to the play total and team sacks get assigned individually based on Quickness. Broderick Thomas is of specific interest to me because in a vacuum his Quickness is the lowest out of that bunch, but RELATIVE to all of the other players in TB's atrocious front seven Thomas's Q is almost double everyone else, so the engine assigns an abnormally-high % of sacks to that particular player.

 

In similar fashion, Q's function for DBs appears to be related to INT YARDS (specifically AVG yards / int). Most DBs on the game hover at about 50Q at average condition. Tyrone Braxton (DEN), Maurice Hurst (NE), Hasty Stargell and Washington (JETS), are a few of the players who have low Q ratings (higher INT than Q) and those players are expected to have low Yards/INT. KC is an interesting team to perform an empirical study on because CB Ross has higher INT than Q and CB Lewis has higher Q than INT, and Cherry has much higher INT and Q than Porter. In a sample size of 100 or 1000 sims I would expect Ross to have higher int volume than Lewis but Lewis would have more yards and TD, and that Cherry would always be better than Porter by a quantifiable amount

 

If I recall correctly, TSB15's programmers actually re-programmed these elements into MAN/COM gameplay rather than just the sim

[Knobbe]

On 8/31/2015 at 4:10 PM, pjfoster13 said:

Quickness is not a USER-functioning attribute but rather an input into the sim itself when you simulate seasons on SKP v SKP. It controls the distribution of Sacks and Interception yards (and TD) under simulated conditions

 

This is not accurate.  Quickness doesn't do anything in SKP vs SKP season simulation.  There are separate defensive sim stats that control this.