More lessons from the data

It turns out that the ShaftLab graph of a golfer's swing is not exactly what the article, nor the ShaftLab literature, implies. Let's see what this means, and use it to better understand when and how the golfer is bending the shaft.

The ShaftLab trace covers a lot more than the downswing

This "lesson" isn't about shafts so much as it is about ShaftLab. The trace isn't exactly what it purports itself to be.

According to the ShaftLab literature and the article, a ShaftLab trace represents the downswing. In some of their writing, they are more careful with their wording. The trace is slightly more than the downswing; they describe it as "from the beginning of load (just before the start of the downswing) to impact." But most of their writing just describes it as the downswing.

In fact, only the last 42%-57% of the trace is downswing! Only about a half of the trace is actually downswing. The rest is backswing; the trace contains almost a third of the backswing.

How did I come to that conclusion? I guess I was primed by the fact that the durations of the trace didn't make much sense. A downswing duration is well known to be in the 200-350 millisecond range, with most professionals in the lower half of that range. But the trace durations for the pros in the ShaftLab manual were listed at 400-600 milliseconds. Something was wrong here. But I didn't bother to dig deeper to find out what it was until September 2017.

 That was when George Hodgetts got into a discussion with me about kick velocity. We still haven't agreed on that issue, but he sent me some very interesting traces as part of the discussion. George has built his own version of ShaftLab, attaching four strain gauges to a shaft in a quadrature configuration -- just like ShaftLab. They are lower on the shaft -- the midpoint rather than just below the handle -- but the idea is the same. George's traces looked nothing like any of the traces I have seen nor measured myself with ShaftLab. How do we explain the difference? After I wrestled with it for a bit, I found a way to look at George's traces in a comparable way to one from ShaftLab. In the picture at the right, we have a single-hump ShaftLab graph on top, and George's graph on the bottom. Here are the things I had to do to get some measure of equivalence: Combine separate traces on one set of axes. Red for lead/lag and dark for toe up/down. Scale them to look more like the ShaftLab traces. George's had both curves with the same maximum deflection of 3 inches. I don't know why it did (doesn't make sense to me), but I halved it for the graph to a maximum of 1.5 inches. That gets it part way to what ShaftLab looks like. While I did that, I also inverted it to the same direction as the ShaftLab curve (that is, lead deflection is positive). I also changed his axis labeling from sample count to seconds, for easy comparison to ShaftLab. But they still don't look quite right. Or they didn't, until George reminded me that his graph was an entire backswing-downswing set; he even sent me a version of the graph where he marked the start of the downswing. Once I saw that: I marked off a section and shaded it in green, corresponding to the ShaftLab trace. If you look at the green-shaded area, it has very much the same shape as the ShaftLab trace. Yes, there are significant differences, but there are substantive differences from trace to trace out of ShaftLab. That is its effectiveness as a fitting tool; it captures differences as well as similarities in swings. But we still have a problem comparing George's traces to ShaftLab's. If a ShaftLab trace corresponds to the downswing, then "Start of Downswing" should be very close to the left edge of the green-shaded area. It isn't. Not even close! Only 57% of the ShaftLab trace consists of downswing and transition. How can that be? One possible explanation might be that a ShaftLab trace starts way before the downswing. That would also be consistent with my earlier observation about the duration of ShaftLab traces: 400-600ms, compared with known downswing durations of 200-350ms. How can we test this hypothesis? I have several measures of my own swing, including a ShaftLab trace and a video that I can step through frame by frame. So I am in a position to compare them in detail. I just never bothered to before. The discussion with George motivated me to. Here is a comparison of  my ShaftLab trace to frame snapshots of the actual video at those moments in time.[1] My swing is closer to a "ramp" rather than a "one-hump" swing, and the conclusion is even more extreme than George's. Only 42% of my ShaftLab trace is downswing; that would be the yellow-shaded portion of the ShaftLab graph. This is pretty strong confirmation of our inference from George's graph. There is nothing particularly odd about either George's swing nor mine. Since we're talking about time durations, we could do a sanity check by comparing both swings against Novosel's "Tour Tempo" hypothesis: ideal swings have a ratio of 3:1 for backswing time vs downswing time. George and I have monitored this for our own swings over the years. He has been stable at 2.7:1 and I have been stable at 2.9:1. So both our swings have timing that is consistent with good golf swings. The issue has to be with ShaftLab's algorithm to determine the start of the backswing; it doesn't work very well. I have gone back and looked at the rest of my conclusions in this article. Only one depends on this flaw in the ShaftLab algorithms: the statement about the time and direction of maximum bending. I have rewritten that part to conform to the new information.

(10) When and how does maximum bend occur?

This is the assertion in Weathers' article. He says,
"For nearly all players, toe-up bend is far greater than any other bend during the swing. It peaks for most players somewhere near the middle of the downswing."

That would imply that the peak bend occurs very close to the peak toe-up bend, and a consequence of that is that the direction of maximum bend is mostly toe-up.

A close examination of the actual data shows no such thing. Fortunately, ShaftLab traces explicitly show the point of maximum bend, as well as its amount. That bend is broken down into toe-up and lag bend, which can be read off the graph. So it is easy to calculate just how far from toe-up the maximum bend is. Here is a plot of the bend components at maximum bend. There are 18 points: the nine tour golfers, each with a driver and a 5-iron.[2] If Weathers' implication is correct, most of the points should be close to the toe-up axis. The graph shows a completely different result.

Yes, there is a toe-up tendency, but only rather slight -- definitely not the dominance that Weathers' statement implies. Consider:
• If the maximum bend were dominated by toe-up bend, then the data points should be clustered next to the "Toe Up" axis of the graph, perhaps no more than 10°-15° from it.
• Such is not the case. The average over all swings has the maximum bend 37° from toe-up. That's a long way from toe-up; a 50-50 split between toe-up and lag would have been 45°, only slightly larger than what we measure.
• A quarter of the swings are closer to lag bend than toe-up bend. (That is, below the 45° line.)
• There are two points that actually show a lead bend component. Let's ignore those -- the red-circled points -- for now, along with the third red-circled point. We'll see later where they come from.
How about the difference between the driver and the iron? Can we say anything about that? Not really. I calculated the mean and standard deviation for each, and they were closer than you would expect even for pure random chance.
 Driver 5-Iron Average 36.6° 37.7° Standard Deviation 18.0° 18.0°
So there is nothing consistent we can say about what direction the shaft bends from golfer to golfer. How about when the shaft bends?

Here is a distribution of the time maximum bend occurs, measured in hundredths of a second before impact. There is certainly some consistency there:
• All but 5 swings see peak bend at .08-.12 seconds before impact, with seven of them (more than a third) smack-dab at 0.10 seconds.
• Of the other five, two of them (at .14 and .20) come from the only single-peak swinger in the sample.
• The remaining three (at .23, .26, and .36) come from a few of the double-peak swings. Specifically, they come from swings where the maximum bend occurs at the earlier peak.

By the way, now we know where those red-circled angles of maximum bend (including two with lead bend) come from. If the maximum bend is in the first peak, it is probably in the backswing. We'll see more about this below.
So the time of maximum bend has a very different sort of result from the very scattered direction of maximum bend. The timing has a strong trend to about 0.1 second before impact, and even the data outliers have an explanation related to the shape of the swing's trace. In the next section, we'll try to understand the different swing shapes and how they influence the peak bend.

Interpretation of swing profiles

I originally referred to this section as "speculation" because I don't have complete data about things like where the transition is for each of the swings in the ShaftLab documentation library. I am also speculating about the cause of the trace shapes. This is the start of a discussion with new information: the evidence that about half the ShaftLab trace shows the backswing, not the downswing. So feel free to take issue with my interpretations. But... if you are challenging my starting premise (that transition is roughly halfway through the trace), you will need better counter-evidence than I have just presented.

First, how many different swing profiles are there, for fitting purposes. The ShaftLab orthodoxy says three: Ramp, Single Peak, and Double Peak. As I went through the traces, there seemed to be a significant number of golfers with a fourth shape, a mix of Ramp and Double Peak. (I'm especially interested in this profile because it describes my own swing.)

But once we correctly identify the transition point, there are only two shapes of downswing: Early Peak and Late Peak. The other differences in the profile are due to differences in the backswing, not the downswing at all. I don't know how much sense it makes to fit a shaft to the backswing; we should be focusing on the downswing behavior.

Let's look at a few traces to see what I mean.[3] I'm using traces from the "library" that is part of the ShaftLab manual. I am identifying transition arbitrarily around the middle of the trace for several reasons:
1. George and I have actual swings where we can tag the transition, and that is where it lies.
2. It gives reasonable downswing durations, unlike the full trace that ShaftLab purports to be the downswing.
3. It lends itself to reasonable descriptions of what is actually bending the shaft during the downswing. (The double-peak downswing doesn't lend itself to any reasonable interpretation that I have ever seen, but I can explain it if only the second peak is during the downswing.)

Let's start with the Single Peak swing. This graph was made by lifting the trace's pixels from the library graph and copying them directly to this graph. The first half of the trace shows not downswing, but the last third or so of the backswing. The trace is a graph of bend in the toe-up/toe-down plane, not the total bend. I haven't shown the lead-lag trace, and it does affect the total bend. But it doesn't dominate either the size or the position of the maximum bend; the toe-plane bend either dominates or shares determining those properties.

So the Single Peak swing has the maximum toe-plane bend near the transition, rather early in the downswing. Even if it isn't exactly at the transition itself, the lead-lag bend can't move it too far away from transition.

What is a good physical explanation of shaft bend in the Single Peak swing? It seems to me that the golfer exerts most of the bending moment on the grip while executing the transition from backswing to through-swing. There is no lightening of the grip to pull the club along the shaft axis. Just set the lag, the plane, and the grip pressure, and turn back then down again.

I have chosen to plot the other three swing shapes together on one graph. They look rather different before the transition. But after transition, they rise (some some need more rise than others) to a late peak, then fall sharply towards impact. This raises the question of what is happening in the swing to make these traces happen.

The thing to realize is the unshaded (backswing) portion of all the traces -- even the Single Peak trace -- consist of toe-up bending due to stopping the backswing in preparation for the downswing.
• The Ramp has a smooth stopping of the backswing, which gradually loads the shaft in a way that increases through the transition and into the downswing.
• The Double Peak stops the backswing more abruptly, bending the shaft a lot. (I'm inclined to think it was a faster backswing to begin with, but have no data to support that guess.) Then the golfer eases off on the bending moment from the hands, in preparation for the downswing. There are a few ways this might be happening, and I would not be surprised if they all contribute to the double peak for some golfer or other:
• A pause in the backswing. Easiest explanation, but probably not the most common in practice.
• Dropping the club "into the slot" -- changing the plane of the swing. I suspect this is common with better golfers.
• Changing from a hand couple to an axial pull. Also likely with better golfers.
• Ungrip-and-regrip at the transition. Perhaps common with less-skilled golfers, where the ungrip-regrip is a well-known flaw.
• The Combination is, as the name suggests, a hybrid of the Ramp and the Double Peak. It starts smoothly like the ramp, then eases off even more. The reasons are probably similar to those for the Double Peak, but from a much lower starting bend.
 With this in mind, let's revisit the distribution of time between peak bend and impact. Here is the same chart, but color-coded with the swing type for each measurement. I see three distinct clusters of data: The red cluster reflects a peak bend during the backswing. It says nothing about the downswing at all. The black cluster comes from the only golfer in the sample with a Single Peak swing. These are early peaks; there is more time left between the peak bend and impact. While this is based on only one sample golfer (two swings) from the library, George Hodgetts' swing, not in the library, is also single-peak, and it has even more pronounced early peak behavior. (George's swing peaks 0.3sec before impact, at the start of the downswing.) In addition, a good look at the nature of the Single Peak swing says that it should produce an earlier maximum bend. All the rest are late peaks, tightly centered around 0.1 seconds before impact. They reflect all the other swing shapes: Ramp, Double, and Combination. In fact, we don't have nearly enough data to see whether there is any tendency of each of them individually. To the level of detail we have, they behave essentially the same. I should note that Pavin's swings are reflected twice in this data. He has two red data points, swings where the maximum bend was in the backswing. But I also found his maximum bend during the downswing, and added those two (one driver and one iron swing) into the data. Those were .10 and .11, smack dab in the middle of the late-peak cluster. If you believe that shaft fitting should be based on the downswing and not the backswing, then the fact of a double peak isn't much of an issue; the first peak occurs in the backswing, and isn't very interesting. From the data I have, the second peak occurs where the Ramp and Combination swings do.So for fitting purposes, you have the Single Peak swing (late peak), and you have everything else (early peak around .10 seconds before impact).. It isn't clear that one would fit an early-peak and a late-peak swing the same; maybe so, maybe not. But even if they are different, that only separates out the Single Peak swingers from everybody else.

Application to shaft fitting

Let's be even more speculative, and imagine what this might mean in terms of shaft fitting. Suppose we start with Wishon's model of shaft fitting. He identifies three criteria -- properties of the swing -- that determine the flex profile you need in a shaft:
• Transition: smooth, average, or forceful.
• Tempo: smooth, average, or fast.
• Release: early, midway, or late.
Assuming he has identified an effective fitting heuristic (and I'll just assume that for now, I won't defend it), our new way of looking at ShaftLab traces might be a very good input to the process, especially if we can guess even somewhat well when the transition occurs in the trace. We know it's somewhere in the middle. With the following interpretation, the exact placement would not be too critical.

• The shaft bend at transition should be a very good proxy for Wishon's "transition" measurement. Note that the slope of the trace is fairly small in the middle of the duration, so any error in estimating the time of transition won't introduce much error in the "transition" measure.
• The time from peak bend to impact should be a very good proxy for Wishon's "release" measurement.
• Wishon's "tempo" measurement is a little more subtle. But reading between the lines, it is a measure of acceleration that feeds into the overall stiffness required of the shaft. So it would be worth trying the maximum bend as a proxy for tempo. In fact, TrueTemper's own experience using ShaftLab as a fitting tool seems to do just that for overall stiffness. They didn't pay as much attention to specific butt and tip stiffness, but our application of the ShaftLab trace to Wishon's process might provide the additional detail required.
We have learned a lot about shaft fitting since Wishon's book was published in 2006, but over the years a lot of successful fittings have been done with this method. It likely gives a very good starting point for fine-tuning the exact shaft selection. And I suspect that the measurement proxies on the ShaftLab trace will give as good a measure as the informal "by eye" descriptions in Wishon's book.

Conclusion

The TrueTemper ShaftLab may be no more, but there were many lessons that can be learned from it and the results it provided to both researchers and fitters. Some of those lessons still need learning by too much of the golf population.

Notes:

1. These are not instrumentation of the same swing, nor were they even taken in the same year. But I have videos and ShaftLab traces taken more years apart than this, and they were substantially the same in shape and duration. So my conclusions here are valid to a first approximation. And they are dramatic enough that a first approximation is all we need.
2. OK, you counted the points. You noticed there are 20, not 18. There are two sets of points for Pavin's swing. If you read ahead, you'll see that his peak bend occurs in the backswing. I also added the points for his swing's peak bend during the downswing.
3. The traces I'm using for each of the shapes are the most nearly representative of each shape, according to my eye.
• Single Peak: Russ Cochran
• Double Peak: Davis Love III
• Ramp: John Cook
• Combination Ramp-Double: Bob Estes