Sidepodcast // All for F1 and F1 for all

Next race: Hungarian Grand Prix, Mogyoród

Selecting the best time to launch an F1 car // Is there a relationship between a car’s launch date and its performance?

Published by Will Davies

In the current F1 landscape, as there is a great focus on testing and time in the car, it is worthwhile considering whether the date a car is launched affects the final performance of the team. Is it vital to a team’s final results for them to launch their cars promptly? And does a late launch leave a team behind, even before the season kicks off for real? Does it work the other way round, or does it not even matter?

In the fall out from the final question of the last High Five Friday, Pat came up with this interesting train of thought:

I think it would be interesting to go back over the last five or six seasons to compare launch dates (or whether the car first ran in test 1 or test 2 or 3) to championship positions - is there a correlation?

Pat was kind enough to already find the majority of the data for me, so it was purely a case of arranging, plotting and analysing it, or at least that’s what I thought. I first decided that it would be most useful to plot on the x-axis the number of days between launch and the first race – we’d then not have to look at 5 distinct data-series on a date-specific scale, thus giving better comparisons. However, the problem here is that different seasons have had different periods of pre-season testing, so the launch windows became separate and awkwardly bunched.

The data

At this point I was plotting against World Constructor Championship finishing position. I felt that it might be a good idea to nondimensionalise the axis so that the points were plotted against a fraction of the time from the first launch to the first race – thereby eliminating that factor of differing calendar schedules. (Nondimensionalisation is essentially the removal of units from a scale and instead the measurement against a relative quantity – one reason being to allow better comparison. For example dividing a time scale by time leaves us with a scale without units.)

To do this I chose to divide the days between car launch and first race by days between first car launch and first race to create a decimal between 0 and 1. The value of 1 representing the date the first car was launched in that season to 0 representing the date of the first race, so the closer to 1 the point is plotted, the nearer the start of the launches-window the car was revealed.

I next felt that plotting against the WCC finishing position was a little too arbitrary a construct, as it did not allow the relative performance of the cars to be represented, so I moved to using the WCC points totals. This proved to be more useful as the difference between cars became more apparent due to the differences in points then being visible. However I still found this change to not be totally satisfying as the points awarded for finishing races has changed over this period.

I chose to get around this not by reapplying a new scoring system (partly for fear it might change the historic results, but also because of laziness) instead I used a similar method to before by dividing the WCC points total by the winning team’s points total that season to generate the fraction of the winning tally. Where 1 represents the winning teams total, etc. Plotting these 2 dimensionless series against each other gives a much better representation of the data, and trends can better be observed.

It is worth noting that the data used does not contain the 3 ‘new’ teams. I feel that including three zero-scoring data points per series will bias the results and therefore might mask more subtle observations – as these teams have been destined to be at the back of the pack regardless of launch date.

Launch date verses championship points
Tap to expand

Year by year

2012: The most recent set of data suggests that the earlier a car was launched then the better it performed. In 2012 the Red Bull, Ferrari, McLaren and Lotus cars were all revealed first (within only a few days of each other) and formed the top of the WCC rankings. The cars towards the back were launched later (although only just), apart from Mercedes who launched comparably late (approximately 3 weeks behind the first teams) but finished mid pack, and usefully don’t fall as an outlier. This positively correlated trend is shown by the gradual slope of the 2012 series (blue).

2011: This year saw a similar state of affairs to that of 2012. The launch-window was even tighter on the most part; there were negligible differences in the teams launch dates, so the 2011 trend (green) displays little correlation. This is apart from Force India who launched late and finished further down the standings, which is the primary reason for the vague correlation that can be observed.

2010: In this year the launches were spread out over a wider time period, which started with Force India, who did so a considerable length of time before anyone else. FIF1 finishing again towards the back has meant that the trend line for 2008 (purple) has formed a negative correlation. It is also worth noting that Red Bull launched last and won the WCC in 2010.

2009: The majority of the points on the 2009 series (red) follow the typical pattern that we might expect (i.e. not 2010); however, there are 3 major outliers that contribute to change in the correlation that can be seen. RedBull launched about 3 weeks after the majority, and came second in the championship, Force India launched almost 3 weeks later and finished ninth, but it was the Brawn car that has the most to answer for. The BGP001 was launched a mere 23 days before the first race and as we all know went on to take the crown. Without this and the Red Bull data points (cars that were seemingly always built to perform well) the trend is closer to the one that Pat first wondered about.

2008: Most of the bigger teams launched towards the start of the window and performed as we’d expect bigger teams to perform and vice versa for the smaller teams. This has resulted in a marked positive correlation in the 2009 (black) series between launch time and result. However, this does not tell the whole story, as there are 2 data-points that outlie so much that I’ve not considered them.

Toro Rosso launched their car 17 days after the start of the season and finished towards the middle of the field, but worse than that is the Force India team who used the 2007 Spyker car until they brought a B-spec model at the Italian GP in September – they scored 0 points. The inclusion of these data points would support the claim that late launches do worse, but as they did so part way through the season, they are invalidated by the use of the old car.

Disclaimer

These data are very specific, and as it is not possible to control for the budget and resources of a team (bigger teams tend to launch earlier) or the skill of a driver in a car (bigger teams tend to have better drivers) it is in turn not possible to declare that any analysis of this is ‘fair’. If it were possible to conduct a suitably randomised study over a long-enough period of time then we’d be able to decide with a lot more accuracy and certainty whether the time between a car’s launch and the first race plays a role in a car’s final performance.

As any mathematician worth their salt knows; correlation does not imply causation. That is to say that just because these two factors look like they affect each other, it does not necessarily mean they do.

Conclusions? What conclusions?

Even with those qualifiers, and when certain outliers are explained / discounted, there is a recurring correlation that does (could) suggest that the earlier a car is launched then the better it will do. However, a better explanation might be that the bigger teams are more likely to finish designing and building their cars earlier and so launch earlier as a consequence. It is also likely that bigger teams like to get their challenger out in the view of the public first so as to get more exposure for their sponsors.

Additionally taking into account that they are more likely to have better drivers, their cars are often more likely to do better – considering that in the past 5 years it has been a big team who has won the WCC, I suggest that the only conclusions we can draw is that the teams that are more likely to win more are the bigger ones (who just happen to be the ones who are also most likely to launch first). Correlation? Only maybe. Causation? No – the only connection is that they are both properties of a bigger team.

Looking at this year’s pre-season situation, the window was small, but the bigger teams still launched at the beginning of it, and I’d suggest they are the ones that are likely (and that we are expecting) to do the best. Williams on the other hand have only just finished their car. I’m not sure anyone is predicting any big jumps in performance from the team and a late launch supports this; as they will be behind in the development of the car. Unless they are particularly hiding anything, but as they’ve only just passed the final crash test this is probably not the case.




  • I can tell just my looking at the points on the graphs that not one could possibly have an r-squared value over 0.9. There's nothing to be learnt from this.

  • I can tell just my looking at the points on the graphs that not one could possibly have an r-squared value over 0.9. There's nothing to be learnt from this.

    not sure i understand that comment? however, not learning something absolutely counts as learning something. #truestory

  • I can tell just my looking at the points on the graphs that not one could possibly have an r-squared value over 0.9. There's nothing to be learnt from this.

    Or you could try being nice.

  • Clearly, as a bear of little brain, the maths here is way beyond me, but I do like the year by year summary. I've been cursing Force India for their late driver announcements and suchlike, but it seems like delaying things has been part of the deal with them that I've just forgotten about!

  • Wow thanks for doing this, Will. This is why I didn't want to take it any further than I did, I couldn't have done this. :) I didn't know about creating unit-less axes and it bends my mind a bit.

    We may not have found a correlation but as Mr C says, that's equally as valid as finding one because now know there likely isn't one - beforehand we had no idea at all. Or if there is one, it is vastly overridden by team resources as Will says.

    That perhaps leads to the question of analysing an individual team rather than a trend, something I tried to do with insufficient data since they appear to launch at broadly the same time. But maybe that is also answered here with the Force India example of launching very late one year and very early the other with no massive difference in result.

    I originally excluded the 'new teams' because I was looking at 2008+2009 and they weren't around then, whereas all the others were, in some form or another - I excluded Toyota for the same reason.

    What kind of magic could Will and Gavin come up with if they worked together I wonder?

  • Ok, to try to get my head around it.

    The y-axis is a teams WCC points divided by the winning teams WCC point total?

    The x-axis is a teams days between launch and first race of a team divided by the days between launch and first race of the team that launched first?

    If I'm understanding correctly do you think there would be any difference if x were the days between launch and first race of a team divided by the days between launch and first race of the WCC team? I assume that would yield some results >1 which I wouldn't know how to interpret if they were worth interpreting.

  • I can tell just my looking at the points on the graphs that not one could possibly have an r-squared value over 0.9. There's nothing to be learnt from this.

    R-squared is a measure of how well points form a trend-line. Having a value near 1 shows there is very little correlation.

    Correlation? Only maybe.

    When we give these series context we can see that bigger teams tend to do better, and bigger teams tend to launch first. So I'm happy about seeing less correlation between what we were comparing :)

  • The y-axis is a teams WCC points divided by the winning teams WCC point total?

    The x-axis is a teams days between launch and first race of a team divided by the days between launch and first race of the team that launched first?

    Correct on both counts. And to answer Christine, the maths isn't *that* hard... Just think of everything having been scaled relative to something useful.

  • Correct on both counts. And to answer Christine, the maths isn't *that* hard... Just think of everything having been scaled relative to something useful.

    Interestingly, I could follow your explanation and conclusions but the graph boggled my marbles. It is usually the other way round :)

  • Interestingly, I could follow your explanation and conclusions but the graph boggled my marbles. It is usually the other way round :)

    It took me a little while to get my head around it. The trick is to focus on one colour (year) and ignore the rest, so looking just at blue for 2012 you have all the datapoints clumped together because practically everyone launched within a few days of each other, except the one way over to the middle which is Mercedes launching much later. The higher the point the better they finished.

  • When we give these series context we can see that bigger teams tend to do better, and bigger teams tend to launch first. So I'm happy about seeing less correlation between what we were comparing :)

    The big teams all launching first is a relatively recent thing. Or at least the leading teams all launching at more or less the same time is a relatively recent thing. If you go back a bit you could guarantee stories in Autosport every year on the relative benefits of launching early versus launching late. Usually it was a comparison between McLaren and Ferrari fighting out the championship despite having opposite launch strategies.

    I wonder if the degree of rule change between seasons would have an effect on the best time to launch. If there is something radically new like KERS or new engines is it better to launch earlier to get a lot of running and data or to spend more time in development to get more of the significant new performance.

  • Another great post from Will. I love how he's looking into something really abstract here!

    As any mathematician worth their salt knows; correlation does not imply causation. That is to say that just because these two factors look like they affect each other, it does not necessarily mean they do..

    May I offer an alternate possibility - that the big teams with lots of money launch earlier, because they have the money and resources to start developing their cars before the 'smaller' teams do?

    Regarding the R-squared comment... data such as these will never lead to a direct correlation so that all the points lie on the R-squared line(R=1). I found in the past that even when comparing data such as laptimes - 60+ data points - there are always outliers from the data that the author requests the reader take into account in their analysis. I have always found that readers are forgiving enough to do so. Besides, I am currently working on a project which assumes R=0.7, which means that some points lie far from the R-squared line. It is entirely reasonable to observe such a value here and it doesn't make Will's analysis any less worthwhile.

    What kind of magic could Will and Gavin come up with if they worked together I wonder?

    I'm keen to collaborate with Will if he wishes to do so, but he's also doing a great job by himself here ;)

  • Regarding the R-squared comment...

    Yeah, I may have got my decimals backwards, closer to one implies more correlation.

    If I'm understanding correctly do you think there would be any difference if x were the days between launch and first race of a team divided by the days between launch and first race of the WCC team? I assume that would yield some results >1 which I wouldn't know how to interpret if they were worth interpreting.

    If I re-jig the spreadsheet you get this: http://t.co/0cnEmJaqOo All it has done is to either stretch or squash the trends as it was only a scaling. The x-axis now can be interpreted as proportion of winning teams time that a team has between their launch and the first race. Ie. in 2009 all teams had longer between launching than Brawn, some even having 3 times as long. The WCC winning teams in each year will now all appear at (1,1).

    The trick is to focus on one colour (year) and ignore the rest...

    This would be a perfect place for one of those click-able SPC-graphs ;)

    May I offer an alternate possibility - that the big teams with lots of money launch earlier, because they have the money and resources to start developing their cars before the 'smaller' teams do?

    I like this interpretation. I wonder how far down the 2014 development path the teams are.

  • I like this interpretation. I wonder how far down the 2014 development path the teams are.

    Joe says Renault had a working engine today and I read somewhere Mercedes do too:

    joesaward.wordpre…y-viry-chatillon/

Comments closed to new entries.