Hierarchy Selection: New team ranking indicators for cyclist multi-stage races

Abstract

In this paper, I report some investigation discussing team selection, whence hierarchy, through ranking indicators, for example when “measuring” professional cyclist team’s “sportive value”, in particular in multistage races. A logical, it seems, constraint is introduced on the riders: they must finish the race. Several new indicators are defined, justified, and compared. These indicators are mainly based on the arriving place of (“the best 3”) riders instead of their time needed for finishing the stage or the race, - as presently classically used. A case study, serving as an illustration containing the necessary ingredients for a wider discussion, is the 2023 Vuelta de San Juan, but without loss of generality. It is shown that the new indicators offer some new viewpoint for distinguishing the ranking through the cumulative sums of the places of riders rather than their finishing times. On the other hand, the indicators indicate a different team hierarchy if only the finishing riders are considered. Some consideration on the “distance” between ranking indicators is presented. Moreover, it is argued that these new ranking indicators should hopefully promote more competitive races, not only till the end of the race, but also until the end of each stage. Generalizations and other applications within operational research topics, like in academia, are suggested.

Keywords: cycling races; dynamics of social systems; hierarchy selection; Kendall $\tau$ rank-order correlation coefficient; ranking teams;

1 Introduction

Operational researchers mostly focus on how to help organisations develop better business systems, to examine how an organisation operates and to suggest more effective ways of working, through individuals and procedures. Let it be called a “microscopic approach” (Ackerman et al., 2018). In contrast, operational research (OR) on teams (scoring and ranking) looks like a “mesoscopic approach” (Corvalan, 2018). That should be part of the modern pillars of OR, within collective choice frameworks research and applications. Indeed, ranking is common in many social life activities: politics, media, economics, academia, - and sports.

It is of common knowledge that for our planet evolution, optimizing (or optimized) selection of choices are mandatory (Lamarck, 1815-1822; Darwin, 1859; Heider, 1958; Ebeling and Feistel, 2011). Both endogenous and exogenous criteria must be provided for optimizing choices among possibilities (Dyer and Miles, 1976; Csató, 2021). Much research has been done on pair competition, as in Verhulst-prone scenarios, through prey-predator or epidemic models (Vitanov et al., 2010; Caram et al., 2015; Saeedian et al., 2017; Gawroński et al., 2022), or for multi-agent interactions (Lambiotte and Ausloos, 2007; Csató, 2020, 2021, 2022). Sometimes, the final choice, when reduced to comparing pairs, lead to paradoxical situations (Condorcet, 1785; Arrow, 1950; Bozóki et al., 2016; Ágoston and Csató, 2022). Moreover, the order of criteria might lead to ambiguities (He and Deng, 2023). In brief, one may recall trivial methods of preference aggregation techniques, like the “means of scores”, - weighted or not, the “Maximum Likelihood Rule” (Le Cam, 1990; Varela and Rotundo, 2016), based on the concept of pairwise preference notions, and TOPSIS, - a method of compensatory aggregation that compares a set of choices, identifying weights for each criterion, normalising scores for each criterion, and calculating the geometric distance between each choice and the ideal one, which is the best score in each criterion (Yoon, 1997; Hwang et al., 1993; Lai et al., 1994).

The discussion, and the subsequent conflict resolution, pertains to a comparison of the evaluation methods, according to criteria (Krawczyk et al., 2019; Krawczyk and Kułakowski, 2021). The drastically annoying deduction seems to stem from the plethora of “preference parameters”, i.e., thus possible criteria. Practically, one turns toward aggregation processes (Munda, 2012), from multi-dimensions toward a single number. The complexity matter further arises when the hierarchy selection is depending both on individual (leadership, PI,…) and on a team (past and present members) achievement scoring.

There are many papers published on “ranking teams”, e.g., among pioneers, Sinuany-Stern (1988), Churilov and Flitman (2006), and Dadelo et al., (2014), although many less than on leaders or athletes rankings. However, an objective “team value” hierarchical scoring is of high relevance for multi-teams collaborations and competitions. Much complexity is known to exist (Fishburn, 1981; Churilov and Flitman, 2006), like in soccer (Ausloos, 2014; Ausloos et al., 2014a, 2014b; Csató, 2020; Ficcadenti et al., 2023); for a recent literature review of contemporary interest on cycling, see Van Bulck et al. (2023). Such considerations are emphasized on one sport event, in the present study, for justifying arguments and subsequent empirical analysis.

The largest amount of papers on team ranking pertains to the most popular sports, of course, like (American) football, soccer, and basketball. Many methods have been proposed. A modern overview by Sorensen (2000) points to various methods for ranking, but mainly pertinent for team duels competitions. Sorensen (2000) and Vaziri et al. (2018) appear to be relevant, but discuss concepts rather than applications. More recent considerations can be found in Csató (2017a, 2017b, 2020, 2021, 2023) and Kondratev et al. (2023). About cycling teams races and subsequent ranking, the literature is much less abundant; yet, recall Van Bulck et al. (2023).

Based on such considerations, some research could be undertaken in finding a new, non classical, way of ranking (professional) cyclist teams. Mutatis mutandis, several aspects can be diverted toward (team) ranking in other social life activities, e.g., in academia or marketing, where some preference scoring, whence some hierarchy, is mandatory. Complex aspects have attracted attention, as in Jose et al. (2008).

Let us specifically consider multi-stage races by professional cyclists. The most famous, prestigious, ones are the “Tour de France”, “Giro d’Italia”, “Vuelta a España”. Beside these “Grand Tours”, there are many others (so called Elite, 2.HC Stage Races, within UCI ProSeries) to which the present considerations apply.

Even though such cyclist races are won by one rider, the role of the team is of crucial importance (Albert, 1991; Mignot, 2015, 2016; Cabaud, 2022). After each stage, a team ranking is provided by the race organisers, according to UCI rules. The teams are ranked according to the aggregated finishing time of the fastest 3 riders of a team for that stage, - excluding all so called bonus time. That sum is cumulated after each stage. At the end of the multi-stage race, each final team time is the result from the sum of such stage times, irrespectively of the involved riders.

Let the finishing time of these 3 fastest riders, $i=1,2,3$, in their arriving order, of team ${(\#)}$, for a stage $s$, be defined as $t^{(\#)}_{i,s}$. In mathematical terms, one calculates the “team (finishing) time on stage $s$” as

$$t_{s}^{(\#)}=\Sigma_{i=1}^{3}\;\;t^{(\#)}_{i,s}\;.$$ (1)

At the end of a $L$ stages race, one obtains each team ${(\#)}$ “finishing time” $T_{L}^{(\#)}$ from the sum of each stage “team time”:

$$T_{L}^{(\#)}=\Sigma_{s=1}^{L}\;\;t_{s}^{(\#)}\;.$$ (2)

Notice that it is often occurring that a race ends with a sprint by a huge bump of riders, thus all such riders are supposedly arriving at “the same time” as the winner. When teams finish with an equal time, they are distinguished, whence ranked, according to the sum of the 3 places of the relevant riders. Let such riders be at place $p^{(\#)}_{i,s}\;$, with $i$=1, 2, 3. Similarly to the above, one can define the “more objective” team ranking place $p_{s}^{(\#)}$ as resulting from

$$p_{s}^{(\#)}=\Sigma_{i=1}^{3}\;\;p^{(\#)}_{i,s}\;,$$ (3)

on stage $s$, and calculate some

$$P_{L}^{(\#)}=\Sigma_{s=1}^{L}\;\;p_{s}^{(\#)}\;.$$ (4)

at the end of the multi-stage race, for the final ranking, - according to the team placing at different stages, again irrespectively of the involved riders.

Notice that both $t$ and $p$ lists do not necessarily give the riders in the same order, due to the last (3) kilometre(s) neutralisation rule, allowing riders to have “technical problems”, tire punctures, even falls, or willingly stop racing, along such a distance.

Arithmetically, within these measures, Eqs.(1)-(4), it can still occur that some teams may have an equal rank at the end of a given stage, - as well as later on at the end of the race, - if the respective sums allow so. (In this study, it is considered that an ex aequo should remained to be what it is, the same rank, without adding a new criterion.)

Two questions seem to arise

• $Q_{1}$: why should one conclude at the end of a multi-stage race that the “final team ranking” results from the time (or place) of riders who do not even finish the race?

• $Q_{2}$: why should one prefer the $T_{L}^{(\#)}$ to the $P_{L}^{(\#)}$ measure?

On $Q_{1}$: it is bizarre that the “final congratulations” about being the “winning team” is based on missing riders at the end of a race. One may admit that some team strategy, based on a rider specific prowess for a given stage, might make sense for specific stages on specific days, but it seems that an objective measure should only consider those riders who finished the race.

For example, admitting this consideration, one avoids a case like that of M. Cipollini in the Tour de France 1999: he won 9 stages, among them were 4 consecutive (“flat”) stages, but he left (abandoned) the race on the next rest day, when stages were reaching mountain climbing features. Yet, his winning times contributed to the final “team total time” measure.

In a sophist way, one could then suggest that a team could bring up new riders in a multi-stage race, replacing athletes at will, as done in basketball, hockey, … or as in other sports involving team duels.

On $Q_{2}$, recall that as far as, e.g., the 1905 Tour de France, the race winner rider was not deduced from the aggregated time-based system, but from a place-based system. This system lasted until the 1913 race, when the time-based system was re-introduced. Even though the change from the original (1903) time-based counting was due to some scandal (the time-based winning rider was accused to have been transported by car or rail during a stage), this change in scoring demonstrates that the rider place-based system can be less easily “adapted” to potential cheating than a time-based system. The more so, thereby I argue, for team based ranking: manipulation, place rigging through trading and/or biasing the final count, seems more easily avoided.

Moreover, one avoids the paradoxical situation, mentioned here above, that, in order to rank teams without ex aequo, one imposes the “place filter” onto the “time filter”. Of course, ex aequo’s can still exist, after summing 3 “small” integers, in Eqs.(3)-(4).

Thus, both for answering $Q1$ and $Q_{2}$, it seems that one can argue that riders should not give up too early in the race, but should keep up in order to be among their team 3 “best” riders, whatever their skill. Indeed, a “good place” on the arrival line is far from having to be disregarded, with these new measures.

In the same line of thought, it is argued that an appropriate measure team’s value, and their ranking, should correspond to the remaining riders of each team at the end of the multi-stage race.

For further numerical display and discussion, it is appropriate to select some case studies; the data used in this study is taken from a recent race, but without any lack of generality.

It is shown and concluded that the new indicators offer some new viewpoint for distinguishing the team rankings through the cumulative sums of the places of riders rather than their finishing times. On the other hand, the indicators indicate a different team hierarchy if the finishing riders are those specifically taken into account. Some consideration on the “distance” between indicators is presented.

The rest of the paper content goes as follows. First, the data used in this study, taken from a recent race, is briefly described in Sec. 2,

In Sec. 3, two indicators are proposed based on the stage finishing time of cyclists finishing a multi-stage race.

In Sec. 4, two indicators are proposed based on the finishing place of cyclists finishing a multi-stage race.

In Sec. 5, two other new indicators are proposed based on the finishing place of cyclists in multi-stage races, bearing a different emphasis on the riders finishing places in the various stages.

In Sections pertaining to some “analysis”, the statistical discussion of results is based on the Kendall-$\tau$ coefficients classically used for comparing ranks in equal size lists (Kendall, 1938; Abdi, 2007; Puka, 2011). Due to a reviewer comment, a brief set of considerations on improving the Kendall $\tau$ usage is found in Appendix A, based on “weighted preferences” notions as discussed by Can (2014).

In Sec. 6, a discussion of the methodology and data analysis is followed by conclusions in Sec. 7; the latter contains a posteriori arguments in favour of the new indicators, with suggestions for further research and applications in organizations confronted to a selection process based on a hierarchy list.

Moreover, in addition, for later discussion here below, it is useful to introduce some notations in order to distinguish individual riders: each is attributed a “bib number” $d$ by the race organizer. Thus, the usually recorded time necessary for the rider $d$ to finish a given stage $s$ is thereafter noted $t_{s}^{(d)}$, while its finishing place on the stage $s$ is called $p_{s}^{(d)}$. When the team and specific rider have both to be emphasized, one will note such measures as $t_{d,s}^{(\#)}$ and $p^{(\#)}_{d,s}$ respectively.

Recall that at the end of the $L$-stage race, riders are hierarchically ranked, according to UCI rules, along their aggregated time, here called $T_{L}^{(d)}=\Sigma_{l=1}^{L}\;\;t_{s}^{(d)}$, in ascending order, which leads to their final (time) ranking. The latter should be distinguished from $P_{L}^{(d)}=\Sigma_{l=1}^{L}\;\;p_{s}^{(d)}$, resulting from merely summing all arriving places of a given $d$ rider in the various $s$ stages.

2 Data

For the present purpose, to deal with data on long (3 week) races would burden the discussion, and would not give much more weight to findings, whence arguments.

Therefore, the case study is the recent 2023 Vuelta Ciclista a la Provincia de San Juan: https://www.vueltaasanjuan.org. The race is part of the UCI ProSeries calendar in category $2.Pro$. The 2023 race, in brief thereafter called VSJ, took place over ($L=$) 7 stages from (Sunday) Jan. 22 till (Sunday) Jan. 29; there was a one day rest on (Thursday) Jan 26. There were ($M=$) 26 teams: 25 with ($n=$) 6 starting riders, and one only with 5 starting riders. Thus, the initial bunch was ($nM-1=$) 155 rider wide; only 133 riders finished the race, due to abandoning or ”arriving after delays” cyclists. There were 7 WorldTeams, 5 ProTeams, 10 Continental teams and 4 National teams. For each team, the official UCI code is hereby used, - for shortening the writing and avoiding undue publicity claims.

The interesting data is obtained from the chronometer officials websites https://www.edosof.com/carrera/517/clasificacion/11#cajaetapas, and similar web sites for the different stages. One can notice that a direct access to the 3rd stage results is missing, i.e., https://www.edosof.com/carrera/517/etapas/31$#$cajaetapas, is empty, but these can be obtained from another media website, like https://www.esciclismo.com/actualidad/carretera/74160.html. The 4th stage data is also misreported, i.e. https://www.edosof.com/carrera/517/etapas/41$#$cajaetapas, but the clumsiness can be immediately resolved, and cross checked through media websites, like https://www.esciclismo.com/actualidad/carretera/74177.html.

A warning: other websites also contain “errors”; for example, not giving the full list of finishing riders. However, any missing information was manually supplemented, and cross checked through different websites. See also some official errors or misprints, described in the Appendix B, but not leading to consequences on the teams place ranking relative values which is the kern of this study.

To be more precise and for completeness, the used data is found in Supplemental Materials.

3 Team Final Time

The classically reported team final time (excluding time bonuses) is $T_{L}^{(\#)}$, as defined in Eq.(2). The $T_{L}^{(\#)}$ of the 26 teams at the end of the VSJ multi-stage race, as so classically determined, is given in Table 1 ; the conventional UCI acronyms are recalled in Table 6.

As explained here above, one can define the thereafter called “adjusted team final time”, $A_{L}^{(\#)}$, calculated as follows

$$A_{L}^{(\#)}=\Sigma_{j=1}^{3}\;\;t_{j,L}^{(\#)}$$ (5)

where, in Eq.(5), $j$ = 1, 2, 3 refers to the “3 best”, whence fastest, riders of the team ${(\#)}$ having completed all $L$ stages. Thus, $A_{L}^{(\#)}$ can only be so obtained at the end of the multistage race. Yet, some possibility exists to “generalize” the concept and its application; see the conclusion Section on “further research” suggestions. Let it be emphasized that these 3 “$j$” riders might be quite different from the various 3 “$i$” riders having contributed to any $t_{s}^{(\#)}$, whence to $T_{L}^{(\#)}$.

The results pertinent for 2023 VSJ are given in Table 1, where one can be comparing $T_{L}^{(\#)}$, not only with the officially published $[T_{L}^{(\#)}]$ results, but also $A_{L}^{(\#)}$; recall to read the comment on data misreporting in the Appendix B, leading to the “definition” of $[T_{L}^{(\#)}]$. Notice at once that the “best team” is markedly different in both measures: IGD is loosing its first rank for MOV, - for about 40 sec. The largest move up concerns TFS; the biggest fall is for TBG. The respective order of other teams is also quite scrambled.

From a statistical difference perspective, the Kendall $\tau$ Rank-Rank correlation coefficient is equal to 0.79692 (Score = 259; Denominator = 325); the 2-sided $p$-value is equal to 0.00.

4 Team Final Place

Similarly to the above, one can define “best team final place” measures, such as $P_{L}^{(\#)}$ together with a $B_{L}^{(\#)}$, based on the final place of 3 “best riders”, at the end of the race: $P_{L}^{(\#)}$ has been defined in Eq.(4); recall that this measure refers to many various riders. In order to adjust the team ranking by only considering the riders ($j$) who finish the race, one defines

$$B_{L}^{(\#)}=\Sigma_{j=1}^{3}\;\;p_{j,L}^{(\#)}$$ (6)

as for $A_{L}^{(\#)}$, defined in Eq.(5). Let it be emphasized again that in Eq.(6), $j$ = 1, 2, 3, refers to 3 “best finally placed” riders, of the team ${(\#)}$ in various stages, but who have completed all $L$ stages.

The $P_{L}^{(\#)}$ and $B_{L}^{(\#)}$ results for VSJ are given in Table 2. Observe at once that the team hierarchy is quite different depending on the measure. The most important scrambling occurs for the main (approximatively 10) teams: in particular, COR moves from the $7th$ rank to the $1st$, TEN gains 3 ranks, but DSM, the leader according to $P_{L}^{(\#)}$, loses one rank, but BOH goes down 3 ranks.

From a statistical difference perspective, the Kendall $\tau$ Rank-Rank correlation coefficient is equal to 0.87692 (Score = 285; Denominator = 325); the 2-sided $p$-value is equal to 0.00.

Furthermore, notice that the $B_{L}^{(\#)}$ values are always greater than the $P_{L}^{(\#)}$ values. This is due to the fact that the finishing riders, after $L$ stages, have not always been involved in the competition for a “good” place at the intermediary stages, while a contrario riders involved in “mountain stages” are not those (usually) involved in “flat stages ending in a sprint”. This again shows that an excellence overall ranking measure for teams demand a consistent competition by all the riders of a given team.

The role of these riders is emphasized through a discussion of the “weight” of riders in contributing to a team “success” in the subsequent section.

5 Overall Best Riders Place Emphasis

It has been discussed that the ranking of a team depends on a specific measure based on a rider place (or time) at the end of a stage. It has been shown how one can rank teams depending on the “best 3 riders” at the end of a stage. Usually, at the end of the $L$-stage race, riders are hierarchically ranked, according to their aggregated time, $T_{L}^{(d)}$, to be distinguished from $P_{L}^{(d)}=\Sigma_{l=1}^{L}\;\;p_{s}^{(d)}$, resulting from summing all arriving places of a given $d$ rider.

In fact, one can similarly calculate for each rider the aggregated final place after the $s$-th stage:

$$P_{s}^{(d)}\;=\;\Sigma_{l=1}^{s}\;\;p_{l}^{(d)}\;,$$ (7)

leading to the $P_{L}^{(d)}$ at the end of the race. Thus, the riders can be ranked in increasing order according to their $P_{L}^{(d)}$ value, called $g_{j,L}^{(\#)}$. The top (e.g., 26, for saving space) best riders according to their $g_{j,L}^{(\#)}$ are given in Table 3 for illustration.

Interestingly, the “winner” of the VSJ race, from such a measure, should then be Tivani, G.N. (from COR); he arrived $12th$ in the time ranking. In contrast, López, M.Á. (from MED) who had the fastest time for the whole race is ranked $15th$ according to the $P_{L}^{(d)}$ measure .

From such a ranking, one can obtain a so called $G_{L}$ team ranking from the finally placed “best 3” riders, i.e.

$$G_{L}^{(\#)}\;=\;\Sigma_{j=1}^{3}\;\;g_{j,L}^{(\#)}\;.$$ (8)

In brief, this “team value” results from the standing places of the best placed 3 riders, in each $s$ stage, but who finished the race.

Moreover, a double summation, on $s$ and on the 3 best three riders who finished the whole race, leads to

$$D_{L}^{(\#)}\;=\Sigma_{i=1}^{3}\;\;\Sigma_{l=1}^{s}\;\;p_{j,s}^{(\#)}\;,$$ (9)

In brief, this “team value” results from the aggregation, over all $L$ stages, of the places of the best placed 3 riders, in a given stage $s$, but for riders who finished the race.

Finally, for comparison of place due to time ranking, the pertinent team ranking based on the final place of the three finishing riders with the best final time, can be calculated as $U_{L}^{(\#)}$; these $U_{L}^{(\#)}$ values and the corresponding team hierarchy are also given in columns 4-5 of Table 3. Notice that one could also calculate the team ranking from the 3 best placed riders in the $P_{L}^{(d_{k})}$ ranking in Table 3, i.e., $\Sigma_{k=1}^{3}P_{d_{k},L}^{(\#)}$, but this is nothing else that $B_{L}^{(\#)}$ given in Table 2.

rank	$T_{L}^{(\#)}$	team	$[T_{L}^{(\#)}]^{(*)}$	team	$A_{L}^{(\#)}$	team
1	77:36:22	IGD	77:05:22	IGD	77:07:53	MOV
2	77:38:46	MOV	77:07:42	MOV	77:08:14	MED
3	77:39:26	MED	77:08:26	MED	77:08:19	IGD
4	77:40:27	SOQ	77:09:27	SOQ	77:12:57	TFS
5	77:41:43	GBF	77:10:43	GBF	77:14:27	AST
6	77:42:24	AST	77:11:24	AST	77:17:04	SOQ
7	77:43:48	TEN	77:12:48	TEN	77:18:03	GBF
8	77:43:54	TFS	77:12:54	TFS	77:18:59	TEN
9	77:46:57	SEP	77:15:57	SEP	77:22:07	DSM
10	77:51:03	BOH	77:20:03	BOH	77:36:01	IPT
11	77:52:22	DSM	77:21:22	DSM	77:38:27	COR
12	78:01:03	COR	77:30:03	COR	77:42:08	BOH
13	78:07:09	IPT	77:36:09	IPT	77:47:28	SEP
14	78:14:34	EOK	77:43:34	EOK	77:52:27	EOK
15	78:27:40	AVF	77:56:40	AVF	78:08:39	EGD
16	78:28:11	ATF	77:57:11	ATF	78:10:40	ATF
17	78:33:26	EGD	78:02:26	EGD	78:15:00	AVF
18	78:49:49	TBG	78:15:02	CHI	78:23:56	CHI
19	78:53:19	PCV	78:18:49	TBG	78:30:46	PCV
20	79:04:23	CTQ	78:22:19	PCV	79:02:34	CTQ
21	79:30:12	CHI	78:33:23	CTQ	79:20:32	ITA
22	79:32:04	EMP	78:52:42	ARG	79:32:09	ARG
23	79:47:37	ITA	79:01:04	EMP	79:32:12	MDR
24	79:48:37	MDR	79:16:37	ITA	79:39:22	EMP
25	79:57:13	ARG	79:17:37	MDR	79:40:03	URU
26	80:05:09	URU	79:34:09	URU	79:45:54	TBG

Table 1: Resulting time ranking of teams at the end of the 2023 VSJ, according to the $T_{L}^{(\#)}$ or $A_{L}^{(\#)}$ indicators, defined in Eq.(2) and Eq.(5), respectively; thus, on one hand, from the (usual) sum of the finishing time of the “best” 3 riders of the team after each stage, and, on the other hand, “adjusted” in order to be only taking into account those riders who finished the whole race, respectively. ^(∗) The central $[T_{L}^{(\#)}]$ data column is the final team time officially reported by the organizers, but including a 31’ error: see Appendix B.

rank	$P_{L}^{(\#)}$	team	$B_{L}^{(\#)}$	team
1	409	DSM	592	COR
2	430	BOH	625	DSM
3	440	GBF	633	TEN
4	462	MOV	635	GBF
5	471	IGD	637	BOH
6	480	TEN	666	MOV
7	547	COR	746	IGD
8	571	SOQ	768	TFS
9	588	AST	872	SOQ
10	612	TFS	879	AST
11	667	IPT	899	IPT
12	882	MED	956	MED
13	979	SEP	1145	SEP
14	1031	EOK	1243	ATF
15	1064	TBG	1330	EGD
16	1119	ATF	1358	EOK
17	1213	EGD	1373	TBG
18	1308	AVF	1444	AVF
19	1354	PCV	1488	CHI
20	1367	CHI	1539	PCV
21	1676	ARG	1777	ARG
22	1690	CTQ	1804	CTQ
23	1831	EMP	1892	ITA
24	1842	ITA	1947	URU
25	1880	URU	1952	EMP
26	1974	MDR	2066	MDR

Table 2: Resulting place ranking of teams at the end of the VSJ, according to the $P_{L}^{(\#)}$ or $B_{L}^{(\#)}$ indicators, Eq.(4) and Eq.(6), respectively; thus, on one hand, from the sum of the finishing place of the “best” 3 riders of the team after each stage, and, on the other hand, “adjusted” in order to be only taking into account those riders who finished the whole race.

rank	$D_{L}^{(\#)}$	team	$U_{L}^{(\#)}$	team	$P_{L}^{(d)}$	$rider$	team	$G_{L}^{(\#)}$	team
1	28	DSM	32	IGD	103	TIVANI	COR	39	BOH
2	31	BOH	34	MOV	139	SAGAN	TEN	40	GBF
3	32	GBF	44	MED	157	GAVIRIA	MOV	40	DSM
4	42	MOV	65	TFS	169	SIMMONS	TFS	42	COR
5	44	IGD	67	SOQ	175	NIZZOLO	IPT	44	TEN
6	44	TEN	76	AST	176	ANDRESEN	DSM	50	MOV
7	47	COR	82	TEN	179	VAN POPPEL	BOH	59	IGD
8	52	SOQ	84	DSM	190	ZANONCELLO	GBF	72	TFS
9	57	AST	86	GBF	194	MARKL	DSM	97	SOQ
10	63	TFS	95	IPT	197	MAGLI	GBF	97	IPT
11	66	IPT	109	BOH	213	RIVERA	IGD	98	AST
12	87	MED	111	COR	213	ROJAS	CHI	110	MED
13	104	SEP	116	SEP	215	GANNA	IGD	140	SEP
14	106	EOK	151	EOK	223	HIGUITA	BOH	166	ATF
15	106	TBG	173	ATF	227	LOPEZ	MED	182	EGD
16	110	ATF	180	EGD	228	PAREDES	SEP	187	TBG
17	119	EGD	185	AVF	232	RUBIO	MOV	191	EOK
18	121	PCV	188	CHI	235	BENNETT	BOH	198	AVF
19	124	AVF	221	PCV	241	OSS	TEN	204	CHI
20	124	CHI	236	CTQ	242	VIVIANI	COR	221	PCV
21	145	ARG	284	EMP	247	KONYCHEV	COR	257	CTQ
22	156	CTQ	286	URU	248	BONILLO	GBF	259	ARG
23	156	ITA	295	MDR	253	CRAS	TEN	277	ITA
24	160	EMP	297	TBG	254	TREJADA	AST	288	EMP
25	164	URU	297	ITA	255	WELSFORD	DSM	291	URU
26	169	MDR	315	ARG	258	MESSINEO	CTQ	306	MDR

Table 3: Ranking of teams after the last VSJ stage, resulting from the sum of places of a team at the end of each stage $s$, i.e., $D_{L}^{(\#)}$, Eq. (9). An other ranking in column 9, $G_{L}^{(\#)}$, i.e., Eq.(8), results from the ranking of the best finishing 3 riders of a team, according to the place they are finishing in each stage $s$. Such top 26 riders, with the sum of their places is given in the central columns, 6-8. The team ranking based on the final place of the three finishing riders with the best final time, $U_{L}^{(\#)}$, is given in columns 4-5.

6 Discussion

From the present research on, and proposal of, new team ranking indicators, one can observe: on one hand, the numerical data much indicates that the new indicators bring new quantitative information on the various team’s “values” at the end of a multi-stage cyclist race. Indeed, it can be observed that the new indicators better distinguish the ranking through the cumulative sums of the places of riders rather than their finishing times. On the other hand, the indicators indicate a different team hierarchy if only the finishing riders are considered.

The new ranking proposes less discussion on the ex aequos. Nevertheless, it is fair to admit that some ex aequos are still possible: indeed, the measures are based on a finite sum of integers. This is unavoidable, but less often occurring if the number of stages $L$ is large. Moreover, ex aequos are less likely if one uses indicators based on the “cumulative weight” of riders as in $P_{L}^{(\#)}$ and $B_{L}^{(\#)}$, as can be seen in Table 2.

A bonus pertains to the methodology: it is simple; it starts from downloading the race place (and time) of each rider after each stage. The final result demands to make sums in an appropriate way. There is no “bonus time” of the finishing riders before further ranking, nor the need to invent complicated arbitrary weights on some inverted finishing race order, as for example done in the “yellow jersey” in VSJ or the “green jersey” competition in Tour de France. Moreover to decouple final race measures from “time or point bonus” due to intermediary sprints, allows for another set of team strategies, as those actually selected by riders or coaches. In this respect, i.e., considering various team best specificities, thus playing on the possible performance of riders according to the stage type, see Rogge et al. (2012).

Thus, the complete list of riders at the end of each stage can be easily downloaded and stored according to their arrival time and/or place. Technically speaking, it is sometimes convenient to organize and to store the lists according to the bib of each rider. Even though some algorithm can be invented, some summations are more conveniently done manually.

For a pertinent comparison of indicators, one has performed a Kendall-$\tau$ correlation test, along https://www.wessa.net/rwasp_kendall.wasp#output. The latter website provides the two-sided $p$-value. Moreover, the latter (free on line) website provides scatter plots of the $X$ and $Y$ variables and alternatively of their respective ranks. For space saving, these plots are not shown, since they are not carrying any peculiar information of present interest. In all cases, the results are found to be statistically significant.

Recall that a positive (negative) $\tau$ indicates a so called high (low) rank correlation. It is found that the number of concordant pairs, i.e., when the rank of the second variable $Y$ is greater than the rank of the former variable $X$, is relatively small., - being even null at the end of this study case.

Remember that it has been pointed out that a brief set of considerations on improving the Kendall $\tau$ usage, in particular within economic perspectives, can be found in Appendix A, based on “weighted preferences” notions as discussed by Can (2014).

7 Conclusions

It is commonly admitted that cyclist races are won by one rider, but the role of the team is of crucial importance (Albert, 1991; Mignot, 2015, 2016; Cabaud, 2022). In fact, cyclist races are quite different from other sport competitions, emphasising individual athletes. Even team competitions, like football (soccer), basketball, hockey, rowing, etc., which sometimes have some focus on specific athletes, or even animals (Lessman et al., 2009) have team quality derived from (integer) numbers, corresponding to some rank and statistics (Anderson, 2015; McHale and Relton, 2018; Kharnat et al., 2020).

Recall that one aim of the study is to propose an objective “team value” measure, with subsequent ordering, whence hierarchy, for multi-teams competitions. Based on such considerations, a new, non classical, way of ranking (professional) cyclist teams is proposed. The numerical analysis is centered on the final hierarchy. Some illustration is based on a recent race, presenting aspects of more famous multi-stage races, but without loss of generality. Thereafter, comparing such new indicators to the usual previous one is in order; one finds distinguishable features, - like clusters, as illustrated in Figs. 1-3, roughly distributed according to the team UCI level, recalled in Table 6. Thus the linear fits are merely a guide for the eye.

The main conclusions are, on one hand, that the new indicators distinguish the team ranking through the cumulative sums of the places of riders rather than their finishing times.The new ranking proposes less discussion on the ex aequos. It is fair to admit that some ex aequos are still possible, since the measures are based on a finite sum of integers. However, this occurrence is less likely when considering place-based ranking.

Moreover, it seems that one can argue that the place accounting should demand more riding action till the end of each stage, thus fuller competition. A strong sport-based argument in favour of the new indicators goes as follows indeed: using the place-based indicators permit to imagine that riders (and team coaches) will have to choose on more different strategies than those existing as of now. Basically, it is expected that riders will attempt to obtain a “good place”, irrespective of their stage time. Thus, no need to say that if such indicators had been implemented, in this 2023 VSJ race, the final results might have been different.

Moreover, these new indicators can be of interest for betting schemes (Yuce, 2021; Etuk et al., 2022), and/or e-gamers (Beliën et al., 2011). However, this study is not intended to predict the results of a race.

As other arguments in favour of the introduction of these new ranking measures, one can consider their interest by sponsors, since the presentation of teams on a podium at the protocol time is the source of a non negligible publicity. Notice that UCI rules permit 6 distinctive jerseys for leading riders in such multi-stage races. There does not seem to be a limit for team “special bibs”. Thus, such indicators can be implemented, thereby increasing the offer to new sponsorship. For example, by extension of the leading rider jersey notion, one could imagine that one defines a “green bib for teams”, similar to the distinctive (say, yellow, in Tour de France) bib, for the best time team ranking after $s$ stages.

In this line of thought, one should remark that the analysis is based on data for which no strategy was a priori developed; the role of a coach is not introduced. It should be relevant, and even exciting to see how new strategies will be developed in order to be the most valuable teams along these new operational lines.

This can be generalized; “further research” suggestions follow.

Indeed, notice that one can calculate intermediary indicators values: e.g., one can calculate for all riders ($i$) their time (excluding or not time bonuses) after $s$ stages

$$t_{i,s}^{(\#)}=\Sigma_{l=1}^{s}\;\;t^{(\#)}_{i,l}\;.$$ (10)

Thereafter the “adjusted team intermediary time”, after $s$ stages, - whence after removing running time contributions from riders not going further than the $s+1$ stage, can be calculated as follows

$$A_{s}^{(\#)}=\Sigma_{j=1}^{3}\;\;t_{j,s}^{(\#)}$$ (11)

where, in Eq. (11), $j$ = 1, 2, 3 refers to the 3 riders of the team ${(\#)}$ having the best finishing time after $s$ stages.

One can also define, calculate, for all riders ($i$) their place after $s$ stages

$$p_{i,s}^{(\#)}=\Sigma_{l=1}^{s}\;\;p^{(\#)}_{i,l}\;,$$ (12)

and

$$B_{s}^{(\#)}=\Sigma_{j=1}^{3}\;\;p_{j,s}^{(\#)}$$ (13)

where, in Eq. (13), $j$ = 1, 2, 3 refers to the 3 riders of the team ${(\#)}$ having the best finishing places after $s$ stages.

Finally, one may hereby propose further research on longer (3 weeks) multi-stage races. One can likely predict that the Kendall-$\tau$ values will tend to become smaller at the end of such races, giving arguments in favour of the application of the new indicators, whence also leading to enhanced competition through new strategies.

Figure 1: Scatter plot of the rank correlation between $T_{L}^{(\#)}$ and $A_{L}^{(\#)}$, with mention of the Kendall $\tau$ value ($\simeq 0.7969$); the best linear fit obeys : y = 0.766154 + 0.943248  x, with $R^{2}$ $\simeq$ 0.88972.

Figure 2: Scatter plot of the rank correlation between $B_{L}^{(\#)}$ and $A_{L}^{(\#)}$, with mention of the Kendall $\tau$ value ($\simeq 0.6246$); the best linear fit obeys : y = 2.473846 + 0.816752  x, with $R^{2}$ $\simeq$ 0.66708. Nevertheless, notice the data points different type of clustering on each side of $r=13$.

Figure 3: Scatter plot of the rank correlation between $B_{L}^{(\#)}$ and $P_{L}^{(\#)}$, with mention of the Kendall $\tau$ value ($\simeq 0.8769$); the best linear fit obeys : y = 0.433846 + 0.967863  x, with $R^{2}$ $\simeq$ 0.93676. Notice the data points disordered cluster below $r=7$.

Appendix A: Improving on the Kendall $\tau$ coefficient: the weighted Kemeny distance.

For examples, in sport ranking, in academia ranking, the disagreement at the top or at the bottom of a ranking might have some drastic influence. In sport, the prize money for the first teams (or riders) is quite weighted. The last teams in the ranking face relegation and loss of money contracts. Thus a swap in positions, in both cases, may be very critical. Thus, the score ruling and aggregation schemes are demanded to be robust and statistically significant (Tsakas, 2020). Therefore, quoting Can (2014) ”it makes sense to assign more dissimilarity (weight) to a change at (…) critical positions.”

As pointed out by Can (2014), Csató (2017b), and/or a reviewer, the Kendall $\tau$ coefficient does not take into account the precise position of dissimilarities when comparing both linear ranking sets. In particular, there is neither a discrimination about the rank difference of a pair in both lists, nor about their relative position in each list. This defect can be (practically) overcome through searching for elementary changes in both linearly ordered ranking, taking into account all possible item permutations. The permutation number values can be ranked in a vector form. Next, one considers sums of $path$ $distances$ between the elementary changes vector components. To find the minimum sum is not a trivial task (Can, 2014); see also https://ncatlab.org/nlab/show/Cayley+distance.

Considering a debatable potential effect on ranking conclusion emerging from a mere Kendall $\tau$ value, Csató (2017b) proposes to weight the position of discordant pairs when manipulating the data in order to obtain two identical lists. He proposes a mere hyperbolic function: $w_{C}=1/r$, $r\;\in\;[1,r_{M}-1]$, based on the lowest rank $r$ of an item of a discordant pair. Conforming to Csató, one can examine if such a weight has an effect when measuring the discrepancy between the resulting indicators discussed the main text. The classical Kendall coefficient $\tau$ would correspond to choose a $w_{K}=1$ for all permutations. For some scientific addition, the permutation values are also weighted through another weight distribution: $w_{A}=\sqrt{1/r}$, smoother than $w_{C}$, as shown in Fig. 4, for the particular case $r_{M}=26$. Thus, one can classically measure the number of discrepancy pairs $D$, obtain the “score” $S$, thereafter the Kendall $\tau$ coefficient from $S/m$, - as recalled in a footnote here above, whence the Kemeny distance $K=N(N-1)(1-\tau)/4$. The procedure can be repeated, appropriately weighting the various swaps, whence obtaining a “weighted Kemeny distance” between pairs, called $A$ and $C$.

	$D$	$S$	$\tau$	$K$	$P$	$A$	$C$	$Q$
$A_{L},\;[T_{L}]$	29	267	0.82154	29	101	29.136	8.9667	27
$A_{L},\;P_{L}$	59	207	0.63692	59	103	31.531	10.493	59
$A_{L},\;T_{L}$	33	259	0.79692	33	131	38.655	12.371	33
$A_{L},\;B_{L}$	61	203	0.62462	61	155	46.351	15.724	61
$P_{L},\;[T_{L}]$	46	233	0.71692	46	152	43.029	13.199	45
$T_{L},\;[T_{L}]$	6	313	0.96308	6	56	17.869	6.2649	7
$B_{L},\;[T_{L}]$	56	213	0.65538	56	126	38.912	13.773	57
$P_{L},\;T_{L}$	48	229	0.70462	48	178	49.409	14.790	48
$P_{L},\;B_{L}$	20	285	0.87692	20	146	42.726	14.199	24
$B_{L},\;T_{L}$	60	205	0.63077	60	140	42.860	14.887	60

Table 4: Characteristics values leading to measures of the “Kemeny distance” (Kendall $\tau$ without normalisation) between pairs of indicators, for a few cases examined in the main text: number of discordant pairs $D$; score $S$; a number of permutations $P$ or $Q$ depending on the chosen vector space (see text); Kendall $\tau$ coefficient; Kemeny distance $K(\equiv D)$ (https://ncatlab.org/nlab/show/Kendall+tau+distance). The weighted Kemeny distance with either square root or mere (-1) hyperbolic weight distribution on permutations is $A$ or $C$, respectively.

This has been done for a few relevant indicators, i.e., measuring the correlation between pairs of indicators due to the number of discordant and concordant pairs (of teams) in Tables 1-3. Not all indicators of the main text are considered, because a few ($D_{L}$, $G_{L}$, and $U_{L}$) contain ex aequos, leading to practical complications, the solution of which would carry too far from the main points. The results are gathered in Table 4.

Notice that practically, in order to compare the ranks of pairs of teams, it is first useful to organize the teams in alphabetical order, giving them the appropriate rank for a given indicator, as summarized in Table 5. Let it be also observed that the number of permutations in order to reconcile two vectors, or, in other words, measuring their distance depends on the order of the axes in the chosen coordinate space. Thus, the number of permutations $P$ starting from the ranking, e.g., in Table 5 differs from the corresponding number of permutations $Q$ to be performed in Tables 1-3; necessarily $P\geq Q$. Alas, the path minimizing metric in order to reconcile two lists is not trivial to find (Can, 2014). The “winners’ decomposition”, in Can (2014) wording is chosen to count the numbers of permutations. Notice that the indicators distances are also necessarily ordered: $K\geq A\geq C$.

These considerations should be further pursued. Indeed, such ranks can be considered as network node degrees, thereby leading to identifying key nodes in a network (Csató, 2017a), - i.e., their $centrality$ (Rotundo, 2011; Negahban et al., 2016; Yazidi et al., 2022). Moreover, the search for the path minimizing function (Can, 2014) is related to searching for specific paths and communities on networks (Festa and Resende, 2009; Varela et al., 2015; Winston and Goldberg, 2022). This is of much interest when communities, whence potential hierarchy, emerge (Eckert and McConnell-Ginet, 1992; Cohendet et al., 2004; Demil and Lecocq, 2006; Yearworth and White, 2018; Wu et al., 2019).

	rank in indicators
$team$	$A_{L}$	$[T_{L}]$	$P_{L}$	$T_{L}$	$B_{L}$
ATF	16	16	16	16	14
ARG	22	22	21	25	21
AST	5	6	9	6	10
AVF	17	15	18	15	18
BOH	12	10	2	10	5
CHI	18	18	20	21	19
COR	11	12	7	12	1
CTQ	20	21	22	20	22
DSM	9	11	1	11	2
EGD	15	17	17	17	15
EMP	24	23	23	22	25
EOK	14	14	14	14	16
GBF	7	5	3	5	4
IGD	3	1	5	1	7
IPT	10	13	11	13	11
ITA	21	24	24	23	23
MDR	23	25	26	24	26
MED	2	3	12	3	12
MOV	1	2	4	2	6
PCV	19	20	19	19	20
SEP	13	9	13	9	13
SOQ	6	4	8	4	9
TBG	26	19	15	18	17
TEN	8	7	6	7	3
TFS	4	8	10	8	8
URU	25	26	25	26	24

Table 5: Alphabetical order of teams with their rank in the studied indicators.

Figure 4: Plot of the weight values distribution used for comparing pair dissimilarities in the indicators discussed in the main text: Csató, Ausloos, and Kendall respectively: $w_{C}\;=1/r$, $w_{A}\;=\sqrt{1/r}$, $w_{K}=1$, where $r$ is the lowest rank of a member of the discordant pair.

Appendix B: Anomalies in 2023 VSJ data reporting

The official data source https://www.vueltaasanjuan.org/clasificaciones/ contains several “errors” (or “misprints”, or “confusions”). In particular,

• The finishing time of the 1st stage winner was (is) reported to be 3h 19’ 36”.

  Since the first 85 riders from the main bunch finished the 1st stage in 3h 19’ 36”, together, - 16 teams with more than 3 riders in the bunch, those teams supposedly final time should be (3 $\times$ 3h 19’ 36” =) 9h 58’ 48”, obviously.

  However, it was reported that the first 16 teams are supposed to have finished in 9h 57’ 48”, on stage 1, leading to a 1’ error.

• For the final stage (stage 7), the winner supposedly rode the stage in 2h 33’ 41”.

  Since three relevant riders for 22 teams finished in a bunch together with the winning rider, their team finishing time should be (2h 33’ 41” $\times$ 3 =) 7h 41’ 03”.

  However, the organisers report these team finishing times to be 7h 11’ 03”, thereby missing 30’.

Thus, the final team stage “overall time” becomes shorter by 31’, for the main teams. For completeness, the official time is given in Table 1, as $[T_{L}^{(\#)}]^{*}$. Fortunately, these errors do not carry over on the place ranking relative values. For completeness, the used data in the main text, for each stage, is found in Supplemental Materials.

Acknowledgements :

to conserve anonymity, in the peer review process, no acknowledgement is hereby presented; it will await publication time ; reviewers, editor, and private communication expert will be mentioned

Data availability : data is freely available, see text.

Funding : none.

Disclosure Statement on competing interest : Neither relevant financial nor non-financial competing interest has to be mentioned.

UCI	Team Sponsor	country
acronym	World Teams (WTT)	acronym
AST	ASTANA QAZAQSTAN TEAM	KAZ
BOH	BORA - HANSGROHE	DEU
DSM	TEAM DSM	NLD
IGD	INEOS GRENADIERS	GBR
MOV	MOVISTAR TEAM	ESP
SOQ	SOUDAL QUICK-STEP	BEL
TFS	TREK-SEGAFREDO	USA
	Pro Teams (PRT)
COR	TEAM CORRATEC	ITA
EOK	EOLO-KOMETA CYCLING TEAM	ITA
GBF	GREEN PROJECT-BARDIANICSF-FAIZANE’	ITA
IPT	ISRAEL - PREMIER TECH	ISR
TEN	TOTAL ENERGIES	FRA
	Continental Teams (CTM)
ATF	AP HOTELS & RESORTS / TAVIRA / SC FARENSE	PRT
AVF	AV FATIMA-SAN JUAN BIKER MOTOS-ELECTRO 3	ARG
CTQ	CHIMBAS TE QUIERO	ARG
EGD	GREMIOS POR EL DEPORTE-YACO	ARG
EMP	EC MUNICIPALIDAD DE POCITO	ARG
MDR	MUNICIPALIDAD DE RAWSON	ARG
MED	TEAM MEDELLIN-EPM	COL
PCV	PANAMA ES CULTURA Y VALORES	PAN
SEP	SINDICATO DE EMPLEADOS PÚBLICOS DE SAN JUAN	ARG
TBG	TEAM BANCO GUAYAQUIL	ECU
	National Teams (NTM)
ARG	ARGENTINA	ARG
CHI	CHILE	CHL
ITA	ITALY	ITA
URU	URUGUAY	URY

Table 6: UCI acronym, team sponsors, and (3-letter country abbreviation ISO-3166-1 ALPHA-3 conventional notation) country registration for competing teams in the 2023 VSJ race; the conventional UCI levels are distinguished; the alphabetical order of UCI acronyms is used.