The Scapegoat
Scapegoats. Every football fan has one. If the manager would just see that Jimmy Spuds is utter shite, and furthermore, would drop him, then there would be nothing keeping the team from European and eventually world glory. This is a theme you’ll hear, with variants, across world football, as much as we Rangers fans think we’re in a unique situation.
I’m fascinated with the notion of scapegoats. I know, you’re rolling your eyes and insisting that Jimmy Spuds really is that pish. But hear me out. Scapegoating is a natural human psychological mechanism, one of the tricks our brain plays on us to keep us alive. It’s the promise of something better than this, a more comfortable life. A lottery win, a pay rise, a holiday, a new boss. We convince ourselves that there’s one thing standing between us and glory, and eliminating that thing will make everything wonderful.
This has its roots in evolutionary psychology. It’s not a completely esoteric concept either as both of my two favourite fantasy/sci-fi humourists touch upon the concept of mental self-preservation in their own way.
Terry Pratchett first, outlining his concept of knurd in ‘Sourcery’.
“In a truly magical universe everything has its opposite. For example, there’s anti-light. That’s not the same as darkness, because darkness is merely the absence of light. Anti-light is what you get if you pass through darkness and out the other side. On the same basis, a state of knurdness isn’t like sobriety. By comparison, sobriety is like having a bath in cotton wool. Knurdness strips away all illusion, all the comforting pink fog in which people normally spend their lives, and lets them see and think clearly for the first time ever. Then, after they’ve screamed a bit, they make sure they never get knurd again.”
Similarly, Douglas Adams had already developed a device known as a ‘somebody else’s problem’ (SEP) field in his book ‘Life, The Universe, and Everything’.
“The Somebody Else’s Problem field… relies on people’s natural predisposition not to see anything they don’t want to, weren’t expecting, or can’t explain. If Effrafax had painted the mountain pink and erected a cheap and simple Somebody Else’s Problem field on it, then people would have walked past the mountain, round it, even over it, and simply never have noticed that the thing was there.”
While knurd and SEP are used as humorous devices, both ideas tap into a very real phenomenon; that human beings live their lives filtering out details they don’t particularly want to deal with.
It’s this mechanism that largely keeps us sane, although Pratchett’s portrayal of Head of the Ankh-Morpork City Watch Samuel Vimes, whose attempts to treat his natural knurdness result in alcoholism, is a largely sensitive rendering of how human beings can lose their way if they find themselves unable to switch off. Both the concepts of knurd and SEP fields relate to the concept of perception; either having too much (knurd), or being happy to write it off for self-preservation (SEP). Each of these ideas are similar to the real life psychological concept of cognitive dissonance, which is an observation that humans are capable of extraordinary mental gymnastics in order to process new information in a way that ensures it’s compatible with existing beliefs. An example might be the individual that posted on a Rangers messageboard that finishing third in the league was actually a good thing, because it meant potentially one extra home Europa League qualifying matches, with attendant additional matchday revenue being generated. (That it probably worked out quite well that way in retrospect is neither here nor there.)
Humans are also group animals, and form very tight knit social groups. This is a part of our anthropological background, where we formed small groups, tribes, clans etc. There have been studies carried out that suggest that humans can’t conceive of more than around 200 individuals; this is the upper limit of our brain’s address book. As such, humans have developed a strong notion of the ingroup and outgroup. The ingroup is everyone in your clique, good wholesome people. The outgroup are bad people who want to steal your crops and burn your land. While the concepts of tribe, clan, and family aren’t as strong as they once were, humans can still determinedly sort themselves into small, like-minded groups. You only have to look at how politically partisan the west has become over the last few years. I’m also reminded of a joke about Rangers’ fans; “if four Rangers fans washed up on a desert island, within a week there’d be 5 supporters’ clubs.”
Psychological tools such as scapegoating, cognitive dissonance and recency bias are probably all coping mechanisms that help us negotiate pitfalls, and probably came about as part of our evolution from primates into space-faring, coffee-drinking, grudge-holding millennials.
It’s no surprise then that humans are mad for a scapegoat, a fall guy, a patsy. Football fans are no different, but for them the scapegoat is usually someone involved in the team or elsewhere at the club. Get rid of that person, and the club will flourish. Sometimes it’s the manager, often it’s a player. But sacking a manager is a lot more complicated and expensive than dropping a poorly-performing player. So some hapless body on the pitch becomes ‘the target of the boo-boys’ as the cliché goes, because like overdraft charges, nothing helps a player low in funds at the confidence bank than taking more off him.
In the summer of 2017, after Martyn Waghorn departed Ibrox for Ipswich, I tweeted “with all of the previous season’s scapegoats gone, who will take up the mantle this time round?” A few names were suggested, but two years on (and with Waghorn having switched to Derby County for a reported fee of £8 million,) it’s interesting to look back at the passing of the baton of the bêtes-noir.
Around this time two years ago, Kenny Miller picked up Rangers’ player of the season award for 2016/17. Shortly afterwards he scored our first European goal in six years. He subsequently departed the club in what tabloids like to describe as ‘acrimonious circumstances’ and he’s a bit of a persona non-grata among large swathes of the support, online at least. While part of this may be down to, to date unproven off-field rumours, the online support had his neck on the block for poor performances as early as the start of September 2017. He was slowing play down, too slow, too old. If only the manager were to drop him…
Lee Wallace also fell victim to this fate. Despite his 250 games for the club, and being the captain until his injury troubles, he’s regularly pilloried online. He didn’t create enough, didn’t score enough, wasn’t a great defender. I’m not saying that either player was in scintillating form, or should have been immune from criticism, but I do find it fascinating how quickly some players become pariahs among their own support. Miller and Wallace lasted perhaps four games in 2017/18 before people were calling for their heads.
Subsequently, the former fell out of favour and the latter missed most of the season through injury. Did the team burst into a scintillating patch of form, cuffing teams left, right and centre? Not significantly. Realistically, once a scapegoat has been removed, a support will target another player, the new one thing standing in the way of glory. Who will that player be? What I’ve noticed is that it’s not necessarily a spell of poor form that identifies a scapegoat. There can be a personal circumstance, coincidentally or not. Wallace was the captain. Miller had a spell at Celtic. Andy Halliday who got it in the neck for most of 2016-17, 2017-18, and bits of 2018-19 was…well, Halliday is a mad Rangers fan which conversely seems to draw more opprobrium his way.
Sometimes the scapegoat will have a character trait that, in the eyes of the fans, is causing his poor form. Carlos Pena struggled with fitness. Ryan Jack used to play for Aberdeen. Graham Dorrans is also a Rangers fan. Not always though. Lee Hodson bucked the trend somewhat, being an inoffensive (literally) player, but after Miller and Wallace were effectively removed from the team in 2017-18 the baton of derision passed first to Josh Windass, then Russell Martin. Windass, similar to Miller, was being slaughtered online while simultaneously winning the SPFL player of the month for February 2018. Martin’s sacrificing was perhaps more symptomatic of the modern era. While not exactly setting the heather alight, he formed a relatively effective partnership with David Bates that saw Rangers concede 13 goals in 10 games, which while not great was as good as any other of the frequent centre-back combinations. (Although going into the last game of the same season, they’d allowed 8 goals in 9 games, 4 of those being penalties, three very soft at that, and recorded 4 shutouts.) A key example of the sort of lazy group thinking that results in people being blamed for things that isn’t their fault is David Bates’ passing ability. Nary would a post on Rangers messageboards or on Twitter pass without someone feeling they had to add a withering, qualifying postscript about the centre-half’s distribution. It was de rigueur. According to Modern Fitba however, Bates’ pass completion rate in 2017-18 was the 11th highest among Scottish Premiership centre-backs at 80%, with the average being 76.03%. Russell Martin was even higher at 82.2%.
Martin had become the most modern of things, a meme scapegoat. A meme scapegoat isn’t much different to a usual scapegoat – a player is targeted for abuse – but the opprobrium is exacerbated by online personalities and the free sharing of various types of content. An echo chamber. By the end of the season, Martin had become the worst player many fans had ever seen, utterly incapable of playing football, wouldn’t get a game in the juniors. A thread on Follow Follow following his release from Norwich City in August 2018 contained the following choice descriptions;
“he’s shyte”,
“as centre back cover…hesabsolutely (sic) howling,”
“a staggeringly shite defender,”
“thought he’d jacked in football years ago,”
“bring him home – home being the fire,”
“No harm to the guy but he’s the worst centre half I’ve ever seen,”
“One of the worst centre half’s we’ve ever had,”
“I thought Dalcio was a terrible footballer but Martin made him look like Maradona,”
“Why is there a thread about a Sunday league standard player?”
“he is utter pish doubt he would get a game for a junior team”
“Fcuk I forgot about Kiernan or I chose to, but he was like Maldini compared to Martin.”
“He’s not cut out for any level.”
“Almost forgot about the penalty a game he use to give away.”
“Nothing against the guy, he’s just an appalling defender who gave me nightmares.”
“He was directly responsible for a goal against in about 90% of the games he played in for Rangers. Seemed a decent human being but he was a truly awful centre half.”
“Karthoarse (sic)”
“Hope he retires and doesn’t make another set of fans suffer him.”
“I think he was in (Celtic’s) top five players for assists for last season…”
“Tried his best for us, unfortunately his best was utter p*sh.”
“Just a shockingly bad footballer. How that guy got to the level he was at I cannot grasp”
“Nothing against him personally but he’s up there with Argyriou, Perry etc for banter years centre half calamities.”
“He was complete dung one of the worst CB’s I’ve ever seen play for us.”
“Undoubtedly one of the worst ,if not THE worst centre halfs (sic) I’ve seen in a Rangers jersey in my life.”
“Crazy he was continually played by Murty despite him costing use several goals!!!”
“The blame lies with the man who signed him, and then continued to play him, even though it was completely obvious that he wasn’t anywhere near good enough.”
All of these comments were posted in little over 24 hours. The usual hyperbole, shot through with the strange sport-flavoured journalese that many football fans seem to write in. It’s also typical of how the reputation of a scapegoat diminishes exponentially further. But more on Mr. Martin later.
At the start of the 2018-19 season legendary ex-Liverpool and England midfielder Steven Gerrard took control of the team. He, with recruitment director Mark Allen, has set about building his own first XI, with Kenny Miller and Russell Martin having left the club, and Lee Wallace nowhere to be seen. Bruno Alves, Harry Forester and Fabio Cardoso have also departed, but while there’s often a sense of optimism at the early part of the season with a new high heid yin in the dugout, there have yet again been some online grumbling at some remaining members of the squad proving that there’s always some player to complain about. Early on in 2018-19 Hodson, and Windass continued to draw ire until each departed, Halliday has provoked sharp intakes of breath whenever he’s named on the team sheet (while Jamie Murphy assumed the mantle of ‘First Team Regular Who Now Nobody Likes For Some Reason’ until his season ending injury.) More recently people were grumbling about Jon Flanagan until he lost a stone in weight and started slide tackling everything that moves. It never stops.
Of course, while just as with politics and science-fiction, there’s a possibility that angry online football fans’ voices are amplified more loudly than others’, it does seem to me that scapegoat culture among Rangers fans has become worse since Twitter went mainstream in 2009. It’s worth noting though that more and more adults have access to the internet, particularly since the iPhone was released in 2007, and Rangers’ financial meltdown of 2012 has probably paid a part, but I think it’s safe to say that for the last six seasons, Rangers fans have had a totem to hang their frustrations on.
Statistics
One man’s Jimmy Spuds is another man’s maligned scapegoat. But with critiques of players’ individual performances being historically in the eye of the beholder, how can we tell if a player’s actually performing badly or not? Enter advanced statistics, the cause of, and solution to, all of life’s subjectivity.
I wrote a couple of years ago about the origins of advanced statistics in sport. Such metrics have been slow to engage the Scottish football fan’s imagination, although that’s partly due to circumstance. Scotland has its own football infrastructure but is part of a United Kingdom with Northern Ireland, Wales, and England, with all the shared services that entails. The latter country dwarfs the other three in terms of population, and the amount of money and marketing invested in the respective countries’ games is even more skewed. As of 2017 Scotland had 10% the population of England, but its combined domestic TV deal is worth just 1.3% of its southern neighbour. In other words, for each £1 the Scottish game brings in from TV revenue, the English earn £7.61. The bottom line is that Scottish (and Irish and Welsh) football plays second fiddle to the English, and that’s also reflected in the availability of statistics. Scotland has suffered over the last 25 years of being the Premier League’s poor cousin, even down to the level of statistics available. When Barcelona and Spain tiki-takaed their way to European football domination in 2008, analysts and tacticians everywhere heralded the importance of pass completion. Ten years on, this basic data is still not readily available to Scottish football fans. Nor are key passes, attempts on target, corners, aerial duels, and crosses, information about the English Premier League I can find out relatively easily.
Indeed, some of the Advanced Scottish Football Statistics (I’m going to use the abbreviation ASFS through this article) guys in Scotland (and I) have created or ‘acquired’ their data from various sources (match footage and BBC text commentaries are two such sources.) New ideas like these can take a while to seep into the public consciousness.
This all leaves the burgeoning advanced football statistics community in Scotland in a bit of strange place, as the nation has lurched from virtually no statistical analysis (it’s difficult to even find reliable information on assists. Assists! They’ve been around as a metric since 1993!) to fairly complex information relating to all facets of a team’s performance.
With the launch of the tactics and analysis site Modern Fitba, a number of guys that have been ploughing their own furrow of Scottish football analysis have come together in one place, using the same basic data to analyse the sport. This is welcome, if only because it’ll hopefully synchronise analyses. In his first post on the new site, the Backpass Rule’s Matt Rhein wondered if the public interest in advanced statistics has plateaued. I’m not sure if it has permanently, but perhaps it just needs a bit of time for public perception to buffer a little.
A further obstacle facing the ASFS community is the lack of statistical literacy in the public at large. Matt has suggested it’s improving, but there are still many people across the UK (and beyond) that I would suggest understand numbers, but who are unaware of such basic statistical concepts such as the difference between median and mean, regression analysis, and the principle that causation does not imply correlation. That’s not to say people are thick, because there are plenty of scientists, journalists, and politicians that don’t appear to understand statistics very well either – probably deliberately most of the time, in fairness.
Sometimes though there seems something underhand about statistical misuse. A couple of years ago I blogged about a viral infographic that stated the number of workers killed in the construction of the World Cup stadia in Qatar had reached over a thousand. Radio 4’s statistical programme ‘More or Less’ also picked up on the unusually high number, and established that the organisation that had produced the infographic had used a figure for all migrant worker deaths in the whole of Qatar since hosting the World Cup has been awarded to the country. (The rates used for other Olympic and World Cup stadia construction deaths were those actually killed while constructing those sports grounds.) Said organisation knew they weren’t using the correct figures, but were unrepentant because it was a political means to an end. Newspapers and social media users didn’t seem to question the infographic either while sharing it. Maybe they didn’t understand it. Perhaps they just wanted to believe it.
“91.7% of all statistics are made up on the spot”, goes the famous quote attributed to, fittingly, many different individuals. Whoever first uttered it, it’s a bon mot that seems to encapsulate the public’s attitude to statistics in general; a mild distrust of their provenance, followed by an inclination to not pay much heed to them.
Whether it’s a deliberate attempt to make as much cash as possible (newspapers) or an attempt to mislead to gain political capital (political causes), I think many people are slightly put off by statistics. There’s also large swathes of right and left-leaning individuals and groups that are disavowing ‘facts’ for feelings, each side arguing that subjective truth is more important than objective analysis.
Of course, there are further sections of the population that will believe any statistic they see, and those that will believe any statistic they see that reinforce their preconceived ideas, ignoring those that don’t. There’s another factor at play here though, and that’s the general statistical illiteracy of the population at large. Statistics is a large and complex branch of mathematics, and while I don’t pretend to fully understand it myself, as a result of years of study and using big numbers in my professional career, I hope I’ve got a decent handle on it.
I think that many people shy away from the more objective approach of statistical analysis precisely for the same reasons they’re drawn to identify a scapegoat. It’s that primal urge inside us all to avoid the things we don’t want to give credence to, to bury the lede, to ignore the monsters under the bed. It’s why people don’t get lumps checked out and ignore letters from the bank.
It’s worth remembering though that while numbers allow us to analyse a subject more passively, we still need to convert numbers to information, and quite often subjectivity sneaks in there. How do we interpret our data? What do the relationships of our data suggest? Have we controlled for all variables? What variables are we going to control for? What are variables? Even after all that, two analysts can both be satisfied that for a model sample size is healthy, variables are controlled for, and regression is healthy, and still interpret the results completely differently. Even people interested in stats aren’t above a bit of confirmation bias – seeking out information that reaffirms what we already believe and ignoring (there’s that word again) that which doesn’t.
All that said, statistics are a useful tool for helping us shed the fog of confirmation bias and try and look at things how they actually are, not how we conceive them (if you want philosophy, you’d be best heading elsewhere.)
In statistical analysis, we hear the term ‘control for variables’ a lot. But what do we mean by it? Let’s look at the example below;
My family are, subjectively speaking, tall. My father and I are both 6’2”, my sister is 5’10”, and my mother is 5’8”. I prefer to use metric units due to imperial being evil.
Ranking:
| Person |
Height (cm) |
Rank |
| Me |
188 |
1 |
| Dad |
188 |
1 |
| Sister |
180 |
3 |
| Mother |
173 |
4 |
(The first statistical error we might fall into is suggesting any findings from this data is representative of the population of Scotland as a whole. It wouldn’t be, as we’re dealing with a tiny sample size (referred to as the ‘P’ value) here.)
We might then want to establish what the average height of people from Scotland is, and this is where we find the first variable we want to control for. Average height data is generally broken down by gender, male and female.
Ranking by gender:
| Person |
Height (cm) |
Gender Height Percentile |
Rank |
| Sister |
180 |
99.24 |
1 |
| Me |
188 |
95.7 |
2 |
| Dad |
188 |
95.7 |
2 |
| Mother |
173 |
93.8 |
4 |
So, controlling for both Scottish nationality and gender, we can see that my sister is actually proportionally the ‘tallest’ member of the family. This is a pretty simple example, but there’s no real limit to the number of variables we can control for to try and analyse data. Having established that my family are tall by international metrics, we can then try and establish why. Ethnicity could be a variable. My paternal great-grandfather was English, so we can then bring in average height data for England. This is slightly higher than for Scotland, which could explain why my dad, sister, and I are proportionally taller than my mother, but doesn’t explain why we’re all taller than the UK average. In fact, looking at English ethnicity might represent a phenomenon known as correlation does not equal causation. This occurs in controlling variables when a variable coincidentally fits with our dataset. It’s tempting to suggest that because X tallies with Y, then X causes Y. This is not necessarily true as the data sets may be corralling and not displaying a direct relationship between each other.
In fact, bringing in my first cousins on either side, and my uncles and aunts, allows us to control for more variables, and decade of birth seems to have an impact as well. My younger maternal first cousin, born in England in 1990, is the tallest member of my family, controlling for gender, ethnicity, location, and decade of birth. My maternal uncle, his father, is one of the shortest. That’s all using a very small sample size though, and that’s a pitfall of statistics. Limited data sources can skew results, so analysts always try to find as much data on the subject as they can.
As per of my analysis above, I postulated that my cousins’ Irish heritage influenced their height. I need to be wary of conflating correlation with causation. This is a phenomenon where one of the variables you’re controlling for matches up incredibly closely with your data set. A famous, fatuous example is the satirical Nicolas Cage films released per year, versus number of drowning deaths in a swimming pool in the US. This has an r2 value of 0.666 suggesting a reasonably strong relationship but it’s clearly an example of correlation, not causation. (Well, as far as we can tell.)
What is an r2 value? It’s very simply the coefficient of determination, and helps us understand the relationship between any independent variable, and its dependent variable. On a scale from 0-1, 0 suggests no relationship, while 1 implies a strong degree of interdependence.
For example, consider goals scored (independent variable) and points won (dependent variable). It’s received wisdom that there’s a connection between scoring more goals and winning more points, and pleasingly we can evidence this using statistics.
As you can see, there’s a strong correlation between goals scored and points won, with an r2 value of 0.94. The same is true of goals conceded; the fewer you let in, the more points you win. Note that while the directions of the trend lines are opposite for each scenario, the r2 value is both strong and positive. The complicated explanation for all this is that r2 measures the distance of each individual value from the mean, but more simply put, r2 represents how closely two sets of data are related. We can therefore comfortably infer from the analysis that the more goals a team scores, and the less they concede, the more points they’re likely to win.
‘Likely’ is the operative word. The r2 value not being 1 suggests there’s not an exact correlation between the two datasets, although they’re pretty damn close.
We could look at another set of data to see if there’s any correlation. Let’s go for latitude of home stadium vs. Points won. With an r2 value of 0.3, we can safely say there’s no direct relationship between latitude and points won. But what if we introduce average attendance as a variable to control for? We now see the r2 value jump up to 0.6.
The goal (ahem) of advanced statistics, in any sport, is to try and pick out patterns, and identify the small margins of error that can be eradicated, prepared for, or removed, in order for a team to improve overall performance. We can also see that sample size, controlling for variables, and regression analysis are key elements of statistical analysis.
The sooking eggs portion of the blog over, let’s start analysing Rangers’ 2017-18 and 2018-19 performances with statistical help.
Defence
We’ll start with the box scores first, or as I like to call them, macro statistics. Many terms in ASFS derive from baseball and while ‘box score’ probably isn’t one that translates directly, I think it’s useful for headline figures. In baseball, the box score is an inning-by-inning breakdown of runs scored by each team in a game, followed by the cumulative total of runs, hits, and errors. The equivalent in football would be the breakdown of possession, shots on and off target, and fouls we see at the end of match reports, plus the score of course, which feeds into the league table.
And the final league table for 2017-18 did not make pleasant reading for Rangers. The club finished third for the second consecutive season since promotion from the Championship, but what’s perhaps more interesting are the Goals For and Goals Against columns. Rangers had the most effective attack in the division, scoring 76 goals in 38 games, three more than champions Celtic. This represented a 35% increase on the preceding campaign. The number of goals conceded was more concerning; 50 in 38 games, the fifth worst in the division, and 13% leakier than the previous campaign. In my blog two years ago, I’d noted how the club conceding 44 goals (1.16 per 90) in 2016/17 represented their 3rd worst defensive performance in the top flight since 1980; 2017/18 was the worst, leaking 1.32 goals per 90.
That’s not entirely unsurprising. When Mark Warburton was appointed manager in 2015, he came with a reputation for focusing on attacking play, not being particularly organised at set-pieces, and for setting up teams that would win games by scoring ‘one more than you’.
| Season |
Club |
League |
Team Rank |
Goals For Rank |
Goals Against Rank |
GD |
| 2013-14 |
Brentford |
League One (England) |
2 |
5 |
2 |
29 |
| 2014-15 |
Brentford |
Championship (England) |
5 |
5 |
= 10 |
19 |
| 2015-16 |
Rangers |
Championship (Scotland) |
1 |
1 |
= 1 |
54 |
| 2016-17* |
Rangers |
Premiership (Scotland) |
3 |
4 |
3 |
6 |
| 2017-18* |
Nottingham Forest |
Championship (England) |
=13 |
=11 |
=20 |
-9 |
*Denotes part seasons. Stats pro-rated to reflect different number of games played by teams in the division at the time of Warburton’s departure.
Despite being in his mid-50s, Mark Warburton has only been in first-team football management since 2013. The respective seasonal rankings of his teams can be found in the table above (I’ve omitted his games in charge of Nottingham Forest in the latter part of the 2016-17 season. We can see that Warburton’s teams tend to have healthy goal differences (2017-18) aside, but also have somewhat porous defences with only his first season at Brentford and his second half season at Rangers appearing on par with the club’s position.
Upon being appointed Rangers’ new manager in 2015, he immediately signed a mostly new defence; goalkeeper Wes Foderingham, right-back James Tavernier, and centre-backs Rob Kiernan and the returning Danny Wilson joined Lee Wallace to make up a new-look back 4. For the first time in many seasons the team were able to field a settled defence, with the first choice back 4 + GK appearing in almost every game.
| Name |
Position |
Starts (36 max possible) |
| Wes Foderingham |
GK |
36 |
| James Tavernier |
RB |
36 |
| Lee Wallace |
LB |
36 |
| Rob Kiernan |
CB |
33 |
| Danny Wilson |
CB |
30 |
Dominic Ball filled in at centre-back for the few games Kiernan and Wilson missed. Despite this continuity, in hindsight, there were statistical markers that suggested the defence wasn’t performing quite as well as it could. I looked at the defensive performances by the Championship winners from 2008 through to 2018, and while it’s not exactly ideal to compare data across seasons, Rangers’ 2016 winning side conceded 10 more in the same number of games as Dundee (2014), Hearts (2015), and Hibs (2017), despite hugely outscoring Dundee and Hibs and nearly matching Hearts. There was something not quite right at the back.
I looked at the defensive records of the winners of the championship over the past ten years, from 2008 through to 2018. To control for fluctuations in the number of goals scored each year, I subtracted the champions’ goals from the total goals scored in the division that season. This means we can see the percentage of goals scored by the other nine clubs against the champions. In 2008-09, St. Johnstone conceded 9.02% of the remaining goals as they won the division. The Remaining Goals % (RG) dropped steadily over the next few season to the roughly 6% conceded by Partick, Dundee, Hearts, and Hibs from 2013 through to 2017. And then there’s the outlier of Rangers in 2015-16 who conceded 8.61% of RG. Indeed, even Ally McCoist’s hugely-maligned 2014-15 Rangers team conceded goals at a slower rate than Warburton’s (0.88 vs. 0.94 goals per game, although McCoist was only in charge for 17 games in his season in the Championship.)
| Season |
Champions |
Divisional Goals |
Opponents’ Goals |
RG |
| 2017/18 |
St. Mirren |
472 |
409 |
8.80% |
| 2016/17 |
Hibernian |
469 |
410 |
6.10% |
| 2015/16 |
Rangers |
483 |
395 |
8.61% |
| 2014/15 |
Hearts |
525 |
429 |
6.06% |
| 2013/14 |
Dundee |
476 |
422 |
6.16% |
| 2012/13 |
Partick |
568 |
492 |
5.69% |
| 2011/12 |
Ross C |
507 |
435 |
7.36% |
| 2010/11 |
Dunfermline |
464 |
398 |
7.79% |
| 2009/10 |
Inverness CT |
462 |
390 |
8.21% |
| 2008/09 |
St. Johnstone |
443 |
388 |
9.02% |
| McCoist 14/15 |
Hearts |
525 |
444.5294 |
7.15% |
Going into the first season back in the top flight, Warburton and his assistant, legendary Rangers and Scotland centre-back David Weir seemed to take cognisance that the defence needed strengthened. As such, in the summer of 2016, they signed goalkeeper Matt Gilks, centre backs Clint Hill and Philippe Senderos, and full back Lee Hodson. They also brought in 19 year old centre back David Bates on a development loan, which would prove to be significant later on.
Despite this, Rangers’ defensive travails continued. 2016-17 was the first 38-game season in 13 attempts a Rangers team had conceded on average more than 1 goal per game. And there’s the problem; 1.16 goals against per game was on par for a third place team and the data tells us that the fewer goals you concede, the more points you win.
Mark Warburton didn’t last to the end of the 2016-17 season, being replaced in the interim by u20s coach Graeme Murty before Pedro Caixinha was appointed in March 2017. The Portuguese was soon faced with a crisis when centre-back Kiernan and left-back Wallace picked up season-ending injuries. With Hill and Senderos also injured or dead or missing or something, Caixinha turned to the u20s and promoted Bates and left-back Myles Beerman to the starting XI for the visit to Kilmarnock at the start of April. Both held their own, and more-or-less stayed in the team for the last 8 games of the season, missing one match each.
In the summer, the manager once again seemed to concede that the defence needed tweaked, and so another wall of defenders arrived. The Portuguese duo of Bruno Alves, who had won the European Championship the summer before, and the young Vitória Setúbal defender Fábio Cardoso arrived in June, while the Welsh international left-back Declan John signed on loan on the last day of the transfer window at the end of August. A specialist defensive midfielder was added in the shape of Aberdeen captain Ryan Jack.
Despite these promising acquisitions the defence did not improve. After the first five games of the season, the team were already conceding goals at a rate of 1.2 per game. The best it got over the course of the whole campaign was 1.08 in November, two matches after Caixinha had been sacked, and again around the turn of the year. After throwing away a single goal lead over Celtic in the March Old Firm game, the team crumbled. In the remaining 8 games of the season they averaged 1.875 goals scored per game, but conceded at the same rate, and they picked up just 1.5 points per game, as opposed to a season average of 1.9.
By the start of 2017-18, only two of Warburton’s back 4 + GK of 2015-16 were still regulars in the first team; Foderingham and Tavernier. Wilson and Kiernan had been sold, and Wallace was apparently long-term injured, but their replacements hadn’t been much better. Rangers conceded goals at an average rate of 1.32 game, worse than the season before. Using the metric of RG%, this works out at 9.56%, the worst since the campaign before Graeme Souness took over. Now there’s an omen. While Warburton and his players have mostly departed the first XI, the spectre of slightly whiffy defending remained. But why exactly were Rangers conceding so many goals?
I’ve been poring over data for the last 12 months, and I still find it a bit of a mystery. Rangers in general don’t concede an alarming number of goals from set-pieces (despite what Follow Follow might tell you.)
What’s clear is that Rangers didn’t have a preferred method of letting in goals in 2017-18; any way an opponent proposed smuggling the ball into the net, the Gers were open to it, although they did concede 22% of goals as a result of simple, low, key passes
I looked at where Rangers conceded goals from, and the assists that set them up, using a methodology from Germany known as the ‘Half Zone’ system. The range of leakiness was quite broad, but focussed through the middle of the pitch.
| Assists |
RW |
RH |
Cent |
LH |
LW |
| Up to Edge Of Box |
1 |
7 |
2 |
4 |
5 |
| Edge of Box to Halfway |
2 |
2 |
11 |
1 |
3 |
| Opponent Half |
1 |
1 |
4 |
0 |
0 |
.
| Goals |
RW |
RH |
Cent |
LH |
LW |
| Up to Edge Of Box |
0 |
1 |
29 |
5 |
0 |
| Edge of Box to Halfway |
0 |
1 |
8 |
0 |
0 |
| Opponent Half |
0 |
0 |
0 |
0 |
0 |
Let’s turn our attention to the goalkeeping situation. Wes Foderingham had been an almost ever-present for Rangers in the league since he signed, starting 100% of games in 2015-16, 97.4% in 16-17, and 86.4% in 17-18. Ostensibly he’s a decent shot-stopper, but conceding 115 goals in 106 games over that period is a little concerning.
I looked at Foderingham’s save percentage. This metric is defined as the number of shots on target that didn’t result in goals divided by the number of shots on target. Foderingham’s save % history is as follows;
| Season |
Save % |
NP Shots on Target Faced/Game |
NPG Conc/Game |
ShutOuts/Game |
| 2015-16 |
71.2% |
|
0.94 |
0.42 |
| 2016-17 |
71% |
3.92 |
1.14 |
0.32 |
| 2017-18 |
66.3% |
3.06 |
1.03 |
0.30 |
| 2018-19 |
83.3% |
1.67 |
0.33 |
0.67 |
For some sort of control, let’s first look at Jak Alnwick’s statistics. Arriving at the club in the winter transfer winter of 16/17, the Englishman has only dislodged his compatriot on 6 occasions in the league, mostly due to injury to the number 1.
| Season |
Save % |
NP Shots on Target Faced/Game |
NPG Conc/Game |
ShutOuts/Game |
| 2016-17 |
83.3% |
6.00 |
1.00 |
0.00 |
| 2017-18 |
66.7% |
6.00 |
2.00 |
0.20 |
And finally, let’s look at Allan McGregor’s rate so far across all his appearances since his return to the club.
| Season |
Save % |
NP Shots on Target Faced/Game |
NPG Conc/Game |
ShutOuts/Game |
| 2018-19 |
75.29% |
2.53 |
0.68 |
0.44 |
The early indications suggest that since taking over in the summer of 2018 Steven Gerrard has fortified the defence; goals of all types conceded dropped from 1.76 in 2016/17 and 1.84 in 2017/18 to just 0.71. That’s less than half as many goals conceded.
That’s at least partly due to the goalkeeper. One of Gerrard’s first signings was to entice Allan McGregor back to Ibrox, and he certainly proved his worth. His save % was higher than Foderingham’s for either of the previous two seasons in the league, and he managed to maintain similarly high levels in Europe.
But was this improvement entirely due to the goalkeeper? When called upon in the league and domestic cups, Foderingham’s average save % was actually higher than McGregor’s, although this is an admittedly small sample size, and generally against weaker opposition. What’s probably more pertinent is the number of shots on target whoever was keeping goal for Rangers faced; they allowed 2.47 non-penalty shots per game in 18/19, down from 3.45 in 17/18. Overall, opponent’s shots over the course of the season were down from 342 to 278. This in turn is due to three key changes Steven Gerrard introduced, which I’ll elaborate on later.
We now return our attention to Russell Martin. Football fans often like to apply hyperbole to players they like and dislike. ‘Bad’ players are said to be so woeful they wouldn’t get a game in amateur football, they are less competent than most supporters’ grandmothers, etc. Everything is their fault. By the end of last season Martin had become that player. But was he receiving justified criticism?
Back in 2014, a Rangers Twitter fan account published a statistic that ‘proved’ Rangers won more games without Ian Black than they did with him. This was widely shared at the time with the clear message that Ian Black was bad and Ally McCoist should feel bad for selecting him for so long.
“Facts are meaningless. You can use facts to prove anything that’s even remotely true.” So observed the sagacious Homer Jay Simpson. As with many of the Simpsons’ finest gags, this is both satire and truism. We’re back to subjective interpretation of data again.
At the time the Rangers fan account tweeted the stat about Ian Black, it was true that Rangers had won more games when he started than when he hadn’t. However those statistics included matches against lower league opposition in the cups. Black had been rested for those games, and Rangers had picked up the wins. But remember that correlation does not prove causation. By the same token, Black started more games against Rangers’ main rivals in the division, Hibs and Hearts. McCoist’s team tended to falter in these games, but it wasn’t necessarily due to Black.
Including cup games against lower league opposition is slightly disingenuous as they’re one-off games with different variables that can skew data. And bearing in mind the odd staggered arrangement of the Scottish Premiership, it would be good if we could analyse defensive players’ performance while taking into account the strength of the opposition. The table below is a very high-level and not-especially snapshot of some Rangers centre-half performances from the last two seasons.
| Forename |
Surname |
Season |
Mins |
Non Pen GP90 |
Assist P90 |
Win % |
Goals For PG |
Goals For Below Team Average |
Goals Against PG |
Goals Against Below Team Average |
| Hill |
Clint |
2016-17 |
1979 |
0.14 |
0.05 |
39% |
1.39 |
-0.08 |
1.17 |
0.02 |
| Bates |
David |
2016-17 |
630 |
0.00 |
0.00 |
71% |
1.71 |
0.24 |
0.71 |
-0.44 |
| Martin |
Russell |
2017-18 |
1350 |
0.07 |
0.00 |
60% |
2.20 |
0.09 |
1.47 |
0.08 |
| Bates |
David |
2017-18 |
1184 |
0.08 |
0.00 |
62% |
2.31 |
0.20 |
1.46 |
0.07 |
Thankfully Modern Fitba have done just that, by evaluating the quality of chance (using Expected Goals [xG] – I have more to say on this later) Scottish Premiership centre-backs allowed in 2017-18, while controlling for strength of opponent (Elo Ranking). This helps take into consideration the Championship group skew, as well as how much more powerful than other Scottish teams the last few years. Looking at their analyses of quality of chance Rangers conceded, weighted by quality of opposition, it strongly implies that Martin didn’t have all that bad a season – certainly not as bad as the level of invective suggests.
It should be remembered that it takes more than one swallow to turn a man into an island. I think that’s how that saying goes anyway. Rangers’ defensive frailties can’t all be the result of one player’s ineffectiveness. Or can it?
We’ve looked at centre-backs, with no clear issue identified. Changing the left-back doesn’t appear to have an impact, but a more skilful guy between the sticks does. There’s one other person who’s been a virtual ever present in that sometimes calamitous defence…
Tavernier
I like to think I’m a little more rational than the average person, and I’m certainly more rational than the average football fan, but I’m by no means as objective as I’d like to think. This is evidenced by the fact I have my own scapegoat of sorts, James Tavernier. I’m not saying he’s a bad player per se; he’s certainly technically gifted and occasionally capable of audacious skill, but his continued ascension in estimation among the Rangers supporters, and managers, is something that bemuses me.
In fact, he’s a player that’s bugged me intensely for about three years now – what’s worse is that he’s played almost every minute of the last three seasons, and was even made team captain for the last ten games of the campaign (made permanent in the summer of 2018 by incoming manager Gerrard.) This gals me, because in my eyes he’s a hopeless defender and overrated going forward. I’m absolutely in the minority there as most of our fans seem to love him, with those critical of him decried as being ‘da(d)s’, the most heinous insult in modern football. While he hasn’t won any individual awards, he has been selected as right back in the SPFL team of the season for both 2017-18 and 2018-19.
It’s almost the opposite of a cult hero, who is someone lionised despite often being awful. Of course, while religion and psychology provide us with the concept of the scapegoat, they also provide the notion of the messiah, the one, the hero. In that respect, I’m not completely surprised that Tavernier is held in such high adulation; attacking flair players normally are more beloved than their defensive counterparts. In addition, there’s more symbolism in Tavenier’s case. Mark Warburton coming into the club was initially seen as signalling the end of the ‘Banter Years’, three seasons of lower league football and infrastructural strife. The sun was shining, Rangers were scoring lots of goals, and in the first home game of the season, against Peterhead in the League Cup, Tavernier did this:
He’d also go on to score an exquisite volley in the final of the Challenge Cup (coincidentally also against Peterhead), a competition only open to non-top flight teams, and one that had eluded Rangers so far. 2015-16 would be the club’s fourth and (hopefully) final chance to win it.
I thought it would be interesting therefore to challenge my own beliefs in this area, those being that I’m a little sceptical about the ultimate usefulness of advanced football statistics, and that James Tavernier Is The One Thing Keeping Rangers from greatness. You see, while I believe I’m rational and not prone to tribal thinking, I regularly flat out disbelieve some ASFS, and have clearly scapegoated Tavernier. So, what does the data tell us? And can we trust it?
Defending
Let’s look at his defensive attributes first. There’s a notion that modern fullbacks are more offensive players than defensive, and as such they don’t possess or need to possess the full defensive toolkit to allow them to prevent goals properly. In 2016-17, some Rangers fans observed that if Tavernier were playing in a team with a decent defensive midfielder, centre back pairing, and right midfielder, he wouldn’t be so exposed. During the course of the 2017-18 season Rangers tried 5 new centre backs, they signed the defensive midfield captain from the second placed team in the Premiership the previous season, and they added the hard-working RM Daniel Candeias as well. Ryan Jack had a bit of a stuttering start to his campaign before he was ruled out of the rest of the season in January. Ross McCrorie played a few games in defensive midfield, and was looking the part, before he too suffered a season ending injury. None of these players appear to help Tavernier’s defending, and we returned to the message boards’ tattoo of ‘if he had better players around him, he’d play better…’ for the second season in a row.
This is partly true of course; any players would play better with better teammates, but I also feel that the ‘modern fullbacks can’t/don’t defend that much’ trope is selectively applied to Tavernier. You don’t hear people brushing off Declan John, Lee Hodson, and Lee Wallace’s (lack of) defensive prowess quite as glibly. Similarly, people are often keen to highlight Kieran Tierney’s defending, and in picking his English team of the season Gary Neville highlighted how much Andy Robertson had improved his ability to keep weans oot a close. In fact, Liverpool manager Jurgen Klopp has often commented on how important it was that Robertson develop his defensive skillset, and has noted the club’s promising young right back, Trent Alexander-Arnold has to improve similarly.
One of my conclusions two years ago was that Rangers were losing a disproportionate amount of goals due to Tavernier switching off while defending. This generally contradicts the conventional wisdom that attack-focussed fullbacks such as Tavernier are vulnerable to conceding goals after being caught upfield, because they were being caught out in defensive positions. Losing goals on the counter wasn’t hugely true for Rangers in 2016-17, and nor was it the case in 2017-18 either.
Considering the common accusation that Rangers lost a number of goals on the break in 2017-18, I looked at how many defenders and attackers were present in the 18 yard zone (imagine the penalty area was extended to each touchline) when assists were played and goal shots taken, ostensibly to see if Rangers were prone to conceding from overloads and counters.
Firstly, Rangers only conceded three goals as a result of losing possession in the opposition half. The data also suggests that while Rangers weren’t really susceptible to overloads, conceding only 5 of 41 or 12% of open play goals where there were more opponents in the 18 yard zone at the time of assist or goal conceded.
They did concede nearly 27% of all open play goals from situations where there were no outfield players (defenders or attackers) in the 18 yard zone when the assist was made. Of these 11 goals, three were as a result of teams dribbling at goal, and three came via low passes from the central zone. These two methods of conceding were common for the club in 2017-18, regardless of the number of players behind the ball.
It’s a strange scenario where we’re tempted to identify a single cause for conceding all these extra goals, but the study of data counsels against it. My belief, over the last three seasons, has been that James Tavernier is the catalyst for much of these bad numbers. I know that I’ve regularly railed against identifying a single point of blame, but I do have some evidence to back this position up.
As Christian Wulff notes in his Modern Fitba post it can be difficult to statistically analyse individual defenders. That’s mainly because advanced stats in sport are best suited to measuring outputs; actual elements of play that have a clear definition and can be categorically identified as completed; goals, passes, throw-ins, crosses etc. It’s no surprise that most advanced stats relate to the ‘goaliness’ of a team – successful passes lead to successful shots lead to goals.
These metrics are on less sure ground when it comes to measuring outcomes, which can be defined as good things players do on the pitch that benefit the team, but which don’t necessarily result in an attempt at goal. Two examples might be Scott Brown’s arseholiness, and David Bates’ KISS approach to defending.
Defensive units tend to operate in the shadowy area of outcomes, which makes it more difficult to qualify their performance, As Christian alludes to in his post. In fact, despite Rangers’ defence being horrific for three seasons in a row, I’ve seen more analysis on how the Gers can improve their attacking prowess, rather than stop the leaks at the back, and that’s because it’s slightly clearer to quantify the impact of passes and goals than it is missed tackles and interceptions.
Fullbacks are a slightly different case. As time progresses they seem to be utilised more as offensive players, and that’s where analysis of their performance is focussed. Christian quotes Lee Dixon saying that 70% of a fullback’s job description is attacking in the modern era, but even if that’s true it still means that 30% of their role is defensive. So how do we measure a fullback’s defensive performance?
Most of the analysis carried out by the ASFS accounts seems to focus on attacking outputs, as I mentioned above. That’s not massively counterintuitive; even in the days of David Robertson, 27(!) years ago now, a Rangers fullback was expected to get forward, overlap, put crosses in, and occasionally score.
But here’s the rub. We can look at a goalkeeper’s save %. We have analyses of the chances allowed by centreback partnerships. Spatial analysis suggests the team weren’t especially susceptible to turnovers or overloads. None of these metrics necessarily explain why Rangers conceded so many goals between 2016 and 2018, so we have to go further and eyeball each individual goal.
I analysed each goal Rangers conceded in the league in between 2016 and 2019 and noted who the defenders and keepers were, what type of assist was involved, and how the goal was finished. I also looked to attribute major and minor errors to each defender on the pitch at the time. I have not included goals from penalties or direct free kicks.
Similar to total goal contribution, I could then add major and minor goal errors and divide the total by the number of minutes played by each defender to give a total error per 90 (MajGEP90 and MinGEP90). I also did this last year, and both season’s results are below. It’s probably worth remembering that this is a fairly subjective exercise, and your interpretation of what constitutes a defender’s mistake is probably different to mine.
2016-17 Individual Defensive Errors
| Player |
Minutes |
Major per 90 |
Minor per 90 |
Total per 90 |
| Tavernier |
3145 |
0.23 |
0.23 |
0.46 |
| D Wilson |
1762 |
0.15 |
0.31 |
0.46 |
| Kiernan |
2130 |
0.30 |
0.13 |
0.42 |
| Hill |
1979 |
0.09 |
0.14 |
0.23 |
| Wallace |
2386 |
0.04 |
0.11 |
0.15 |
|
Average |
0.16 |
0.19 |
0.35 |
2017-18 Individual Defensive Errors
| Player |
Minutes |
Major per 90 |
Minor per 90 |
Total per 90 |
| Hodson |
393 |
0.23 |
0.69 |
0.92 |
| Cardoso |
929 |
0.39 |
0.19 |
0.58 |
| Martin |
1350 |
0.13 |
0.40 |
0.53 |
| Wallace |
372 |
0.24 |
0.24 |
0.48 |
| Tavernier |
3358 |
0.24 |
0.24 |
0.48 |
| Alves |
1558 |
0.29 |
0.12 |
0.40 |
| Ross McCrorie |
1717 |
0.10 |
0.26 |
0.37 |
| D Wilson |
1078 |
0.00 |
0.33 |
0.33 |
| John |
2284 |
0.04 |
0.12 |
0.16 |
| Bates |
1184 |
0.00 |
0.08 |
0.08 |
|
Average |
0.17 |
0.27 |
0.43 |
2018-19 Individual Defensive Errors
| Player |
Minutes |
Major per 90 |
Minor per 90 |
Total per 90 |
| McAuley |
555 |
0.00 |
0.32 |
0.32 |
| Tavernier |
3234 |
0.11 |
0.14 |
0.25 |
| Halliday |
1561 |
0.06 |
0.12 |
0.17 |
| Worrall |
1840 |
0.05 |
0.10 |
0.15 |
| Barisic |
1240 |
0.07 |
0.07 |
0.15 |
| Katic |
1421 |
0.06 |
0.06 |
0.13 |
| Goldson |
3031 |
0.09 |
0.03 |
0.12 |
| McGregor |
3059 |
0.00 |
0.09 |
0.09 |
| Flanagan |
1185 |
0.00 |
0.08 |
0.08 |
|
Average |
0.05 |
0.11 |
0.16 |
There are a couple of things to note here; firstly, Rangers used a lot more defenders in 2017-18 than they did in 2016-17 and 2018-19. While it’s a mantra online that Tavernier’s defending is continually improving, his MajGEP90 has increased as a proportion slightly from season to season, although his MinGEP90 went down. What’s also interesting is to look at his stats in relation to his left fullback partner. In both seasons Tavernier’s error per 90 has been far higher than his colleague’s, whether that was Lee Wallace or Declan John, Borna Barisic, or Andy Halliday. Now, statistics can sometimes be counter-intuitive; on the surface they might suggest one thing, but on further investigation the opposite turns out to be true. For instance, Tavernier’s error stats might be higher than the left backs’ because he’s having to deal with crosses they’ve allowed, while he’s kept crosses in from his side to a minimum.
I’m not so sure that’s the case here, but let’s look at it a little deeper. Rangers conceded 12 of 41 goals in 2016/17 from crosses, or 29%. 7 came from the left, and 5 from the right. However, of the 7 crosses from the left, only 4 were allowed when John was playing, with the remaining 3 split equally between Lee Wallace (2) and Andy Halliday (1).
But as I’ve hopefully hammered home, football is a far more fluid sport than that. At the end of the 2017-18 season, Dougie Wright carried out a similar exercise, and noted that Rangers conceded more goals from the left than the right. This was picked up by the Rangers podcast Heart and Hand, and began to filter through social media, with the general interpretation being that this meant Tavernier was twice as good at defending as John. This signal boost from internet influencers helps contribute to what I referred to as ‘meme scapegoatery’ earlier in this piece.
Analysing goals conceded on where crosses originated from is a slightly reductive line of thinking. For instance, let’s look at the second goal Rangers conceded on the final day of the 2017-18 season, away to Hibernian. While a two dimensional map would suggest the assist came in from Rangers’ left, and Halliday might be apportioned blame for not preventing the cross, the truth is a little more complex. Consider the set up to the goal. First Hibs took a quick throw-in on the right hand side. Russell Martin, the right-side centre-back had to go over to pick up the Hibs forward. As such, David Bates, the left-side centre-back was drawn to the right hand side of the box to cover, and Andy Halliday tucked in from left-back. Hibs got a cross in, Halliday was beaten in the air, and the knockdown was converted by Scott Allan. While the assist for the goal came from the left, it was actually in the left centre-half zone, rather than the left wing, and none of the Rangers midfielders picked up on Allan’s run.
What’s also quite interesting, while looking at the goals Tavernier was at least partly culpable in the concession of in 2017-18 is that you only see his name for 8 of the first 34 goals Rangers conceded in the league (23.5%), then 10 of the remaining 16 (62.5%.) It’s fascinating to speculate how this freefall in form came about. There are a number of potential reasons; he’d been linked with a transfer away from the club in January, which didn’t materialise. He then agreed a new contract, and was given the team captaincy in February. I think this added responsibility was preying on his mind.
What I suspect happened defensively to Rangers that season is down to a number of factors. Firstly, losing the first and second choice defensive midfielders probably made Rangers a bit more porous in the middle third. Declan John was then injured, with the less defensively inclined Andy Halliday filling in. Finally, towards the end of the campaign, Daniel Candeias, who had been performing so diligently ahead of Tavernier on the right flank, lost his mojo. Tavernier was then more exposed, against better teams, and his defensive inabilities were brought to the forefront. Most footballers will play better with better players around them, but you have to wonder that if he needs so many players around him playing well to look competent, whether he’s worth the hassle.
But all of that doesn’t really matter, does it, because his attacking output outweighs any defensive mistakes he makes, doesn’t it?
Doesn’t it?
Attacking Prowess Myth?
I see a lot of tweets and posts about Tavernier’s offensive attributes; his ‘goal threat’. His delivery from crosses. Bookmakers seem to regularly tweet about his goals and assists contribution from set pieces;
“Rangers captain James Tavernier has scored 29 goals and registered 43 assists since he joined the club. Not bad for a right-back.”
This was posted on the 19th August 2018; it’s the sort of stat that you’ll see shared and retweeted and gain credence among those people that really like Tavernier, and while technically not incorrect, it should be caveated. Most data analysts would granulise the data for a better understanding of what’s going on. Over the course of the season those 29 goals and 43 assists increased to 43 and 57, broken down as follows;
What we can see straight away is that a third of his total goals and around half his assists came in one season in the Championship three years ago. Back then, Rangers played in the lower-division Challenge Cup, and now they’re competing in the Europa League. So let’s do what almost all football analysis is based on, and look at league competition only.
| Season |
League Goals (Pens) |
League Assists |
| 2015-16 (Championship) |
10 (0) |
19 |
| 2016-17 (Premiership) |
1 (0) |
5 |
| 2017-18 (Premiership) |
8 (4) |
7 |
| 2018-19 (Premiership) |
14 (11) |
14 |
| Total |
33 (15) |
45 |
33 goals and 45 assists. Still not too shabby for a right back. We’ll now look at the number of appearances racked up to get these goals and assists, using the ‘Per 90’ method. This is done to try and standardise the data between someone who starts and then plays full matches on a regular basis compared with someone who maybe comes off the substitute’s bench. As such, we’ll also discount penalties, as reasonably standard practice, as they don’t necessarily reflect a player’s goalscoring prowess.
| Season |
Minutes (Possible) |
League Non-Penalty Goals Per 90 |
League Assists Per 90 |
| 2015-16 (Championship) |
3213 (3240) |
0.28 |
0.53 |
| 2016-17 (Premiership) |
3145 (3420) |
0.03 |
0.14 |
| 2017-18 (Premiership) |
3358 (3420) |
0.08 |
0.19 |
| 2018-19 (Premiership) |
3234 (3420) |
0.08 |
0.39 |
| Total |
12950 (13500) |
0.13 |
0.28 |
Broken down like that, we can see two things; one, that Tavernier’s not really a goalscorer. A third of his goals came in the Championship, and another half have been penalties. In terms of his assists, 2018-19 saw him almost get back to his Championship level of creativity. More on that later.
Furthermore, this data now lets us compare and contrast Tavernier’s performance more accurately against those of his team-mates.
| Season |
League Non-Penalty Goals Per 90 |
Club Seasonal Best NPGP90 (min 500 minutes) |
League Assists Per 90 |
Club Seasonal Best AP90 (min 500 minutes) |
| 2015-16 (Championship) |
0.28 |
0.66 (Forrester) |
0.53 |
0.64 (Oduwa) |
| 2016-17 (Premiership) |
0.03 |
0.44 (Dodoo) |
0.14 |
0.36 (Windass) |
| 2017-18 (Premiership) |
0.08 |
0.50 (Morelos) |
0.19 |
0.44 (Murphy) |
| 2018-19 (Premiership) |
0.08 |
0.73 (Morelos) |
0.39 |
0.43 (Middleton) |
His non-penalty goals per game rate was 0.08 in both 2017-18 and 2018-19, which isn’t very high; mildly above average for a full back in the SPFL. Far from being a goal-threat, this meant he was slightly more likely to score than Jason Holt or Ross McCrorie, and less likely than Connor Goldson or Glen Kamara.
Nor was his 2017-18 assist per 90 rate particularly stellar (0.19), although his second assist metric was decent at 0.29 (the pass that sets up the assist). These were roughly on par with his stats for 2016-17 – 0.03 NPGP90 and 0.14 AP90. Last season (2018-19) his assists went up but his 2nd assist rate has dropped remarkably, to 0.08.
As noted above, the full-back enjoyed a goal-filled season in the Championship in 2015-16, with around half of all his non-penalty goals for Rangers coming in that 50-match campaign. It’s worth bearing in mind that his goals dried up a little after that, and particularly his open play goals. As of the summer of 2019, Tavernier has scored 12 goals from open play in the league for Rangers, with 58% of those coming in that Championship season. It’s only since he’s taken over the mantle of penalty taker that he’s really started scoring regularly again. Of his 23 top-flight league goals for Rangers since 2016, 15 have been penalties.
We don’t tend to describe a player who sources most of his goals from penalties as a goal-threat as by using that logic we’d have to consider Steve Bruce and Rogério Ceni as dangerous attacking players. What we can do here is look at Tavernier’s underlying shooting statistics here.
| Season |
Shot Accuracy |
Club Seasonal Average Shooting Accuracy |
Fenwick Adjusted Shot Accuracy |
Club Seasonal Average Fenwick Adjusted Shot Accuracy |
| 2016-17 (Premiership) |
20% |
38.75% |
25% |
49.78% |
| 2017-18 (Premiership) |
33.33% |
34.32% |
41.67% |
43.54% |
| 2018-19 (Premiership) |
42.42% |
41.2% |
48.28% |
53.73% |
What’s apparent is that his shooting was awful in 2016-17, however you cut it. It did improve substantially the following season and this to slightly below average, which would still be difficult to describe as ‘great’. In comparison, the top 5 Rangers scorers for each season, be they absolute, or per 90, tended to have higher than average shot accuracy rates. As would probably be expected; that’s how averages work.
So he’s not big on scoring goals, but he’s creative, right? As noted in the table above, his assists per 90 haven’t been consistently impressive either, outside of the Championship season and 2018-19. However, many members of the ASFS community aren’t fans of using assists and second assists as statistics because they argue it skews more credit to the goal scorer for converting the chance than it does the assistor for creating it. This isn’t entirely unreasonable; think of all the gilt-edged chances your team’s striker misses per season – these had to be put on a plate for him to make a meal of. Instead, Expected Assists (xA) has gained traction among the stats community as a metric for analysing how creative a player is.
(What exactly is xA? It’s a value between 0-1 assigned to a pass depending on how likely the chance it leads to results in a goal. If you think this sounds like the same metric as Expected Goals (xG), you wouldn’t be far wrong; it’s essentially exactly the same. There do appear to be a couple of different variants of xA though. One version only credits passes that led to a shot with the respective xG value of where the shooting player received the pass. Another credits all passes with an xG value, regardless of where they were received.)
I’ve been debating long and hard as to whether I should go on a long anti-xG/xA screed again, but I don’t think I will. They’re still not metrics I’m fond of for various reasons. They measure potential outcomes rather than actual outcomes, they’re derived from data from other leagues on occasion, each model is different, each model only factors in a select number of the mind-boggling variables that influence whether a shot on target goes in or not, I’m not sure they control for goalkeepers’ positioning, and aggregating xG seems counterproductive. Some statisticians will note that xG is flawed, but argue until we develop something more automated and able to factor in all variables, it’s the best we have and we might as well get on board with it. I’m not so sure; I’d actually prefer a simpler xG model, used only for high level shot evaluation.
However, in terms of xA, I’m even less clear on what it actually measures. It’s purported to represent a player’s ability to play a pass that results in a ‘high quality’ shot at goal, but if observed assists reflect the striker’s ability to finish a key pass, then doesn’t xA also reflect their ability to be in the right place to receive a key pass? And similarly to xG, when aggregated it can be difficult to distinguish between high volumes of poor passes and more modest amounts of very good passes.
The relationship between xA and oA (observed assists) is something I pondered in depth while researching this piece, and that’s probably why I’ve become so frustrated about it. That’s partly due to the lack of available data in the Scottish game; there is a potential avenue to raw information, but I feel a bit uncomfortable about the caveats involved in requesting access to it. As such, I have to base my analyses on second hand data, which is a little restrictive.
For example, the Rangers Report has previously provided information via Twitter on Rangers players’ performance across various metrics, which is helpful, but having access to more statistics would make my life easier. Candeias’ xA was twice as high as Tavernier’s, but he didn’t record twice as many assists. More granular detail might explain why that is. For example, when I looked at the relationship between Rangers players’ xA and oA, I noticed that the more central players had a far higher ratio than the wide players; an average of 116.9% for the former and only 88% for the latter. Now, oA doesn’t always correlate very closely with xA, but for Rangers things were quite consistent, and I suspect is due to crossing being a far riskier method of delivery than more central passes.
We should talk about crossing; it’s another of Tavernier’s outputs that appears higher in quality than it actually is. On the face of it, he does put in a fair number of delicious crosses that don’t seem to result in goals; a look at any minute by minute match report will call out 3-4 examples per game of Tavernier putting in a tantalising cross that Windass or Morelos just wasn’t able to get on the end of.
Crossing the ball into the penalty area for a forward to attack with his head is something of a throwback to an earlier period of football, but despite the infatuation with possession play over the last ten years, a good hump into the penalty area is still something European teams try a lot (and something Rangers fans at least long to see.)
Despite that, crossing remains a fairly ineffective method of scoring goals – it’s much better at conceding possession. This Opta piece by Garry Gelade observes that only 1 in 92 crosses results directly in a goal, although you can improve those odds with judicious aiming of your centres. Gelade broke down the success rate of various different types of crosses and found that the most successful type of centre was a ‘driven pass that just crossed the centre line by no more than five metres.’ We’ll come back to this later.
Gelade also highlights further categories of crossing, and their relative success rates. One such type is ‘chipped to the far post originating at least five metres from the touch line and end up no more than five metres from the centre line’. They are further categorised by where they terminate; crosses ending up behind the penalty spot have a 2% conversion rate, while those in front of the penalty spot have a 5.8% success rate.
Like David Beckham 20 years ago, a cross delivered from deep behind the penalty spot is the type of cross Tavernier tends to prefer delivering. It looks good and will probably draw a tantalised ‘oooh!’ from the crowd, but it’s still not necessarily a high quality chance; it’s probably in the no-man’s land between the keeper being able to field it, and the striker just not being able to reach it in time.
Of Rangers’ 67 goals in 2017-18 that weren’t penalties or direct free-kicks, headers only accounted for 18.64%. By the same measure, only 18.5% of top scorers Morelos and Windass’ 27 goals came from headers. In terms of assistors, 30% of Candeias’ assists led to headed goals, 14% of Murphy’s, 14% of Windass’, and none of Morelos’. In contrast, 71% of Tavernier’s assists resulted in headed goals. A similar story occurred in the 2016-17 season, with 19.6% of goals coming from headers, and Tavernier racking up 40% of his assists supplying headed goals.
This trend continued into the 2018-19 season. Of 52 assisted goals that weren’t penalties or direct free kicks, headers made up 15.38%, roughly the same number of goals that were assisted by crossing. While other creative players like Candeias, Arfield, Kent, and Morelos recorded their assists by setting up shots with passes (1 cross, 19 passes, no headers,) Tavernier logged 4 crosses, 3 set pieces, and 4 headers.
| Season (Non Penalty/DFK Goals) |
Cross (%) |
Pass (%) |
Corner (%) |
Free Kick (%) |
No Assist (%) |
| 2016-17 (55) |
13 (23.64%) |
24 (43.64%) |
2 (4.65%) |
2 (4.65%) |
12 (21.82%) |
| 2017-18 (67) |
14 (21%) |
42 (63% |
0 (0%) |
2 (3%) |
8 (12%) |
| 2018-19 (66) |
7 (10.61%) |
39 (59.09%) |
2 (3.03%) |
4 (6.06%) |
14 (21.21%) |
| Season (Assisted Goals) |
Cross (%) |
Pass (%) |
Set Pieces (%) |
| 2016-17 (43) |
13 (30.23%) |
24 (55.81%) |
4 (9.35) |
| 2017-18 (59) |
14 (23.73%) |
42 (71.19%) |
2 (3.39%) |
| 2018-19 (52) |
7 (13.46%) |
39 (75%) |
6 (11.54%) |
| Season (Open Play, Assisted) |
Shot |
Header |
| 2016-17 |
32 (74.42%) |
11 (25.58%) |
| 2017-18 |
48 (81.36%) |
11 (18.64%) |
| 2018-19 |
44 (84.62%) |
8 (15.38%) |
Rangers’ goals by type, 2016-19
| Assist Type |
Cross |
Corner |
Free Kick |
Pass |
| 2016-17 |
1 (20%) |
0 |
1 (20%) |
3 (60%) |
| 2017-18 |
6 (85.71%) |
0 |
0 |
1 (14.29%) |
| 2018-19 |
4 (28.57%) |
0 |
3 (21.43%) |
7 (50%) |
James Tavernier’s assists by type, 2016-19
So what’s going on here? Given the statistics above, one might argue that Tavernier’s crossing is so sumptuous, he’s providing far more than the expected headed goals. The alternate argument, and the one I think is more likely, is that the majority of Tavernier’s crosses are of a type that can only be converted by heading the ball, i.e. 6-7 feet off the ground, thrown into the edge of the six yard box. This is his preferred delivery, but is a more difficult chance to convert, which is why his assist per 90 rate between 2016 and 2018 was so low (relatively speaking) despite the perception he puts a lot of high quality crosses into the box
As I argued earlier, xA measures not only the assistor’s pass but the striker’s ability to find the optimum location and space for a shot. Different attacking players have different strengths; some are good in the air, others prefer tap-ins. Tavernier seemed to have a decent rate of crossing to Alfredo Morelos at least (again, more on this later,) but less so Josh Windass, Rangers’ top two scorers in 2017/18. The reason for this was due to pass selection; Tavernier was continuing to play in his long, deep cross to the middle of the penalty area regardless of who was in the box to get on the end of it. Sometimes I think he puts crosses in without even checking to see if there’s anyone in the box first. It’s then up to the attacker to maximise the potential of the cross. Compare this approach with that of Daniel Candeias, who has (or at least had) a far more considered, intelligent approach to cross selection. He is more likely to play flat, low driven balls into the penalty area. More importantly, he passes to the runs of the attackers already in the area, rather than making them work to get on the end of his passes. Look at his assists for Windass’ goals against Aberdeen and Partick in 2017-18, and his pass to Morelos against Spartak in 2018-19; low, accurate crosses to feet. This is partly why Candeias’ assist per 90 rate (and xA per 90 rate for that matter) has tended to be higher than Tavernier’s, despite both of them operating in much the same area of the pitch most of the time. This is a bit of a metaphor for Tavernier’s overall game; superficially very appealing, but a little one-dimensional when you scratch the surface.
As I have mentioned earlier, there is a knack and a science to delivering an effective cross. I also feel there’s a distinction to be made between a cross, an elevated, curling pass with a zonal target, and a square ball, a pass delivered low from the wings into the centre. Gelade mentions in his piece that ‘the most successful category of cross was a driven pass that just crossed the centre line by no more than five metres.’ This is intuitive; attempting to convert a low cross is arguably easier than trying to get a head or an acrobatic volley on a lofted cross. It’s also worth noting that Gelade’s research suggests that Tavernier’s preferred method of crossing (from deep, swung into the area,) has among the lowest conversion rates.
But what’s the beef? 17 goals and 20 assists from right back in a season is an unarguable return. Right? Right? Despite being involved in 37 of Rangers’ 115 goals last season, unanimous praise is not forthcoming – at least not outwith Ibrox. He was selected as right-back in the SPFL team of the season two seasons running, but in fairness the competition for that position isn’t that strong.
But why do I and his fellow players seemingly not rate Tavernier particularly highly? Why are Rangers fans exasperated that Kieran Tierney is being linked with a £25m move to Arsenal while no-one seems particularly keen to pay silly money for the Gers’ captain?
I think it’s partly due to people that aren’t associated with the club don’t really buy into the ‘17 goals from right back!’ given that 14 have been penalties – penalties are roughly seven times more likely to be converted than any other type of shot, so they’re not necessarily indicative of a goalscorer. Rangers fans will point out that a goal’s a goal, and that’s true.., just because Tavernier’s the preferred taker right now doesn’t mean he’s the only player in the squad with that skill. Defoe has taken a number of penalties in his career, as has Dorrans, Davis, Murphy, and Morelos, though none have quite the same success rate as the right back. And then there are the assists…
‘But Jay,’ I hear you say. ‘You’ve already told us he’s racking up big actual numbers.’
This is true. Well, his assist rate has improved dramatically, and xA and key passes per 90 are also high. Second assists and non-penalty goals remain the same however. And as Yer Boy will tell you, high actual numbers are not necessarily reflective of high level performance. However, there’s a slight risk of metrics and ‘box’ statistics causing a false correlation.
But whatever our opinion of Mr. Tavernier, we can’t deny that his assist rate doubled in 2018-19. What we can look at are the reasons why it did, and if it’s sustainable. And in order to do so, we have to consider how Rangers have set up differently under Gerrard to the last two managers and caretaker Graeme Murty.
Tavernier was signed for Rangers by Mark Warburton. The English coach has a reputation for signing players he’s worked with in the past, and while this isn’t strictly true of Tavernier, the full back had been signed for Wigan by Uwe Rosler. This is significant because when the German was manager of Brentford, his sporting director was…Warburton.
It’s reasonable to surmise that Warburton had identified Tavernier as a potential signing when Rosler was still at Brentford. In many ways the full back fitted the Magic Hat requirement list; young, skilful, not entirely encumbered with tactical nous. Warburton’s approach to coaching is very simple; play short passes until you score. That’s it; no back up, which probably explains the defensive statistics in the previous section. While the Englishman got Rangers promotion back to the Scottish top flight, his tenure didn’t last much longer once there, and he was technically-not-sacked and eventually replaced by Pedro Caixinha.
The Portuguese was a rum sort, and Rangers’ fortunes didn’t improve much under his tenure. Famously he once gave no instructions to the team ahead of the second half of a match against Motherwell, resulting in a chaotic shambles. He was sacked. In the intervening periods Graeme Murty filled the chair as caretaker, and while he’s done a decent job with the club’s development teams, this generally requires following the ethos set out by the club’s sporting director, rather than forging one’s own footballing philosophy. Hence, it was only really the summer of 2018, and with the arrival of Steven Gerrard, that Rangers’ tactics began to evolve.
When analysing how Gerrard has changed Rangers style of play, we should look at how his former employers, Liverpool, set up. As mentioned above, most big European teams try to maintain the same system and playing style throughout the club, from schoolboys to first team. This continuity allows for the clear lines of progression to the first XI for young players as well as allowing for the adaptability and attention to detail elite modern football requires.
As such, under the German Jürgen Klopp, Liverpool have tended to play variations on a 4-3-3 formation, characterised by his use of gegenpressing and ‘Chaos Theory’. For Klopp’s Liverpool, attack is the best form of defence; as soon as the team lose possession, the wide attackers are charged with harrying the opponent to win it back immediately, or force them down blind alleys, ideally out to the wings where there are fewer options to successfully pass the ball out of danger. This is the gegenpressing element.
The midfield three are not necessarily expected to create as much, but to provide a narrow belt in the middle of the pitch to break up play and again force the ball wide. The two wingbacks are charged with pushing forward into the spaces exposed by the narrow midfield and the roving wide forwards, and deliver crosses into the penalty area. This is the chaos theory part. And it was an approach that appeared to bear fruit for the Reds with the fullback partnership of Robertson and Alexander-Arnold racking up 30 assists between them in 90 matches in all competitions.
As soon as Gerrard was appointed his recruitment suggested he wanted to emulate his former employers, to an extent. Not only did he bring along several of Anfield’s backroom staff, he secured loans for the Reds’ Academy players Ryan Kent and Ovie Ejaria, as well as signing fullback Jon Flanagan on a free. Elsewhere, left-back Declan John was surprisingly sold to Swansea just six months after signing permanently. He had been supplanted a few days earlier when the club had completed the signing of the Croatian left-back Borna Barisic. In retrospect, these dealings make more sense; John is more of a dribbler whereas Gerrard presumably wanted a left-wing back that would deliver quality crosses into the box.
Barisic hasn’t really worked out so well so far, and Gerrard ended up rotating him, Jon Flanagan, and Andy Halliday on the left. But things were looking up for Tavernier on the opposite side of the pitch; he was officially made captain and would go on to start all but two of Rangers’ 60 games. As I’m sure you’ve observed, Tavernier’s assist ratio isn’t dissimilar to the Liverpool wing-backs’.
The Rangers Report recently posted a piece about key passes, those passes that set up a scoring chance. A scoring chance is a shot at goal taken in the rectangle formed by the goalposts, the goal line, and the penalty line. Efforts from here are three times more likely to go in than shots from elsewhere.
In their post, Jason noted that Tavernier was the most creative player in the league, averaging 0.89 key passes per 90. Extrapolating out, this tallies 32 KPs over the course of the season. If we multiply this by the 29% likelihood of a shot from here resulting in a goal, then we can reasonably say that Tavernier should have expected to assist 9 goals in the league. Given that he’s on 14, the argument can be made that he’s delivered above and beyond.
But wait, the Rangers Report has more data. 26 of those chances were from open play, and 6 were from set pieces (corners and free kicks.) These result in expected assists of roughly 7 and 2 respectively. His actual assists are 11 and 3. This is where the role Tavernier performs for Rangers is similar to Trent Alexander-Arnold. Not only are both players encouraged to put crosses into the box, both take a large number of free kicks and corners for their teams.
By the middle of March, Tavernier had attempted 152 crosses, 33 more than the next closest player (who happened to be Candeias.) This equates to roughly 5 per match. We can say with some comfort that around 1.5% of crosses in the U.K. result in a goal. Therefore Tavernier should have 2 crossing assists; he had four. So he was assisting more than expected from crosses, open play passing, and free kicks. That’s good, right?
This is where crosses exist, in the hinterland between box scores and xA, with advanced statisticians cast as theoretical scientists, trying to unite classical and quantum physics. And this is also where metrics fall down a little for me because, on the face of it, the data say he’s been delivering a lot of quality service into the box from crosses, passes, free kicks and corners. Analysts will point to xA and scoring chance key passes, but while many football metrics purport to identify quality, they tend to reflect volume instead. Thus if Tavernier has attempted lots of crosses, and completed a portion, he’ll rack up some xA points even if they’re not particularly good crosses. Meanwhile, the ‘key passes’ metric will measure completed passes to the scoring zone, but won’t necessarily measure the quality. Conversely, xA will give us a rough approximation of quality, but normally at the expense of detail. For example, player X might have 5 assists from an xA of 5.2. This looks okay at first glance, but with the average xG of a shot being around 0.10, we’d expect a player to reach 5.2xA after around 52 shot assists. He may have attempted 87 passes to reach this total. This is why I advocate xG and xA being divided by the number of attempts to try and achieve a more instructive number.
However, I reiterate; without access to the raw data, I’m limited as to how I can break down the data, so my only recourse was to look at all 82 goals Rangers scored in the league in 2018-19 and see where Tavernier’s assists fit in.
As previously mentioned, his crossing assist stats are unusually high, but then we can safely assume he’s put in more crosses than anyone else in the league. His free kick assists numbers are high, but he’s taken more free kicks than anyone else in the Rangers team (and by extension probably the league.)
While I disagree with some of the findings of the ASFS community, I do agree with them on many other things. Repeatability is one. This relates to the statistics underpinning a certain metric: basically how much you can be certain that, say, crossing, will deliver a guaranteed number of goals each game or season? This removes the element of chance that assists introduce; a midfielder could create 20 great chances across a season, but if his striker only converts 1 of them, who’s the numpty?
We know that crossing is one of the least likely types of attack to result in a goal, so unusually high numbers here mean one of three things; you have excellent crossers, a striker that’s good at converting, or you’ve been a bit lucky. As none of Rangers’ attackers are particularly effective at heading the ball, this suggests Tavernier falls under categories 1 or 3. And we do know that he attempts a lot of crosses…
What’s interesting to observe is that in the first half of the season Tavernier was assisting from open play at a rate of 0.38/90. In the second half this halved to 0.19. At the same time Rangers’ team goals scored increased from 2.05/90 to 2.29. Why might this be?
The answer, perhaps, is Alfredo Morelos. The Colombian had come to Rangers during the 2017 close season straight from a Finnish summer season and effectively played 18 months without a break. By the summer of 2018 though he was rested and ready to properly take up his role as key to Rangers’ attack. In his first season he scored 14 goals in the league, a tally he would beat by four during 2018-19.
However, the South American’s issues with discipline are well documented; he was sent off a remarkable four times in the league (with one rescinded,) and while his market value continued to grow, Gerrard had publicly expressed his frustration.
Anticipating either Morelos’ leaving in the summer, or just being constantly suspended, in January Gerrard signed his ex-England teammate Jermain Defoe on loan. This marquee signing did necessitate a change in approach; the 5’7” Defoe is unlikely to dominate aerially. Indeed, according to the Premier League website, less than 1% of his English top flight goals had come via headers. (As if to evidence how chaotic crossing is, Defoe headed home a Tavernier cross against Hamilton shortly after signing.) As such, Gerrard appeared to slowly change his formation and attacking approach, reducing the volume of crosses and encouraging Scott Arfield to link up more with Defoe in an attacking sense.
Due to suspension and Defoe’s form, Morelos featured less in the second half of the season. As a result, he only scored half as many times in the back 17 games. This is about the same factor by which Tavernier’s assists dropped. Coincidence? It’s interesting to note that by dint of the positions he takes up, Morelos could be described as more of an old-fashioned inside right, rather than a centre-forward, and I believe heatmaps would illustrate this. Thankfully, The Backpass Rule produced a shot map for Morelos, which does appear to indicate that the majority of Morelos’ shots come from the right-hand side of the goal.
This correlates nicely with where Morelos’ Rangers goals have come from. 21 in the centre of the goal, 9 from the right half space, and only 2 from the left half space. So in the ongoing xA battle, is it Tavernier’s assists that generating Morelos’ goals, or vice versa?
Well, that’s fairly straightforward to construct a guess. Morelos’ GP90 remained roughly the same throughout the season; 0.63 for the first half, 0.55 in the second. Taking this into consideration, and where his goals were scored from, it’s clear to me that Tavernier has benefitted from the fact Morelos prefers to operate in the inside right channel, slightly ahead of him. Because Morelos is so skilled at using his strength, skill, and shooting prowess, Tavernier has seen the benefit with a number of reasonable quality passes being converted into assists. In this situation, it’s the striker driving the xA numbers, not the assistor. There’s also an argument to be made that with most football teams playing a right-footed player at left centre-back, there’s an inherent weakness in the inside right channel for teams to exploit that doesn’t necessarily exist to the same extent on the opposite side.
From here we can eyeball Tavernier’s assists to see if my theories run true and they were abnormally high. For a start, 29% of his assists were converted by headers. This is pretty high; even controlling for penalties, only 11.43% of Rangers’ goals in 2018-19 came from the head. 29% is up there with St. Mirren and Sam Cosgrove.
5 of his assists, or 36%, were converted by Morelos, rising to 50% if we count cup goals. Morelos isn’t as dependent on Tavernier though, with the right-back only assisting 28% of the Colombian’s league goals (33% in all competitions.) This got me wondering; there were a couple of occasions where Tavernier bagged an assist for a goal that in all truthfulness was down to the striker’s endeavours. Morelos against Kilmarnock in the cup, and in the 1-1 draw at Ibrox. Kent against Celtic. While I don’t have access to an xA model, I could note the number of touches each scorer required to score their goal. Kent’s against Celtic needed five. If Tavernier were getting lucky with a number of lower-quality passes, then the number of touches required to finish from his passes might be higher than average.
| Player |
Assists |
Touches to Finish (T2F) |
Average |
| Davis |
2 |
2 |
1.00 |
| Defoe |
2 |
2 |
1.00 |
| Grezda |
1 |
1 |
1.00 |
| Kamara |
2 |
2 |
1.00 |
| Middleton |
3 |
3 |
1.00 |
| Candeias |
5 |
6 |
1.20 |
| Morelos |
4 |
5 |
1.25 |
| Arfield |
7 |
11 |
1.57 |
| Team |
52 |
90 |
1.73 |
| Tavernier |
14 |
27 |
1.93 |
| Barisic |
2 |
4 |
2.00 |
| Halliday |
1 |
2 |
2.00 |
| Murphy |
1 |
2 |
2.00 |
| Kent |
4 |
9 |
2.25 |
| Jack |
3 |
7 |
2.33 |
| Ejaria |
1 |
7 |
7.00 |
This might not be the most scientific method, and to be honest just because an assist requires more than 2 touches to convert doesn’t mean it’s a lower quality pass. That said, this does support my observation that more than almost any player, Tavernier benefited from having a higher class of player directly in front of him to convert his passes. Case in point, his average T2F for Morelos alone was even higher, at 2.2, while Morelos’ needed 1.81 touches to score goals from other sources. Excluding Tavernier, Morelos’ T2F decreases to 1.63.
(Curiously, he’s managed 5 open play assists from his last 24 games, having managed 14 in his 36 starts before that. Is this regression, Rangers’ subtly changing their style of play, or a mixture of both?)
There’s another aspect of Tavernier’s data we should consider; he’s very rarely injured or rested. In fact, if we look at the table above, we can see that he’s played roughly 96% of all league minutes over the last four seasons. That’s pretty impressive, whichever way you slice it, but it’s more notable after Gerrard takes over, because the Englishman has utilised a squad rotation approach. In 2018-19, only two outfield players started more than 30 out of the 38 league matches; Goldson and Tavernier. While he started more games than any other assister last season, the right-back completed more 90s as well, averaging 87.41 minutes per appearance.
| Row Labels |
Average Time of Assist |
Average Rank of Assist |
Mins per Appearance |
| Arfield |
54.86 |
2.86 |
80.55 |
| Barisic |
33.50 |
2.50 |
77.50 |
| Candeias |
42.00 |
2.60 |
68.64 |
| Davis |
27.00 |
1.00 |
63.64 |
| Defoe |
70.50 |
3.50 |
55.59 |
| Ejaria |
68.00 |
6.00 |
72.50 |
| Grezda |
59.00 |
4.00 |
50.38 |
| Halliday |
24.00 |
3.00 |
70.95 |
| Jack |
62.67 |
3.67 |
80.43 |
| Kamara |
39.00 |
1.50 |
81.69 |
| Kent |
23.25 |
1.25 |
77.15 |
| Middleton |
41.00 |
1.67 |
41.73 |
| Morelos |
34.00 |
1.50 |
73.70 |
| Murphy |
14.00 |
1.00 |
80.00 |
| Tavernier |
50.71 |
2.14 |
87.41 |
| Team |
42.30 |
2.20 |
|
It’s worth remembering at this stage that more goals are scored in the second halves of football matches than the first. Similarly, more goals tend to be scored the later matches progress. This appears to be standard across the world, for intuitive reasons; the longer a match goes on, the more likely a mistake will occur due to mental and physical fatigue, and sheer desperation at times.
Rangers’ 2018-19 season was slightly different; the Gers only scored 39% of their goals in the second half, although they still managed 14.49% of their NPGs after the 81st minute. In contrast, Tavernier’s assists tended to come later in games, with 3 of 14 coming in the last 10 minutes (4 in the last 11, 5 in the last 15.) This is another area where Tavernier’s output statistics got a little boost; he was on the pitch longer than every other creative player, and in more games. In short, if Arfield, Candeias, Kent, and Morelos hadn’t been substituted off as often as they had (between them, and including sendings-off, the four failed to complete 45 of their 119 combined appearances,) their own goals and appearances statistics may have ended up more impressive.
Finally, and most subjectively, a number of the assists Tavernier recorded last season were as a result of luck: St. Mirren away, Motherwell home, Hearts away, Hamilton away, Dundee home, Kilmarnock home. He racked up 6 assists in those games that owed more than a little to luck, refereeing decisions, bad defending, and team-mates’ ability. “The harder I practice, the luckier I get,” said some golfer. The more passes Tavernier attempted in 2018-19, the more assists he seemed to rack up. But it doesn’t necessarily tally that he’ll repeat that feat in 2019-20.
Mentality & Positioning
The final area of his game I want to look at is his positioning, which again was something I called out two years ago. While he’s sometimes accused of being caught upfield in an attacking mode, the club don’t tend to lose a lot of goals due to his offensive sensibilities. He does however have a habit of going walkabout in an attacking sense. Most modern full-backs do in fairness, but I rarely see it occur as often as it does with Tavernier. One wonders if it’s a tactical decision, and his team-mates are briefed before the game because I know I wouldn’t be happy if I were playing someone who continually drifted as erratically about the pitch as Tavernier does. I mentioned this in my blog last year as well, how his wanderlust requires other players to adapt their positioning to cover his runs, lest the team be left vulnerable. Essentially this means a centre-back, holding midfielder, winger, and centre forward have to adjust where they are to protect the defence. This abuse of resources would be fine if the team was getting a lot of output from Tavernier’s raiding, but five open play goals in 111 league games isn’t great.
I think his positioning is one of the reasons for him making so many defensive errors. Because he doesn’t think like a defender, he doesn’t really want to be defending, and doesn’t prioritise defending. While he’s actually quite good at one-on-ones when he wants to, I feel he’s often distracted by daydreaming about attacking. When Rangers are transitioning from offensive to attacking phases, Tavernier remains in an attacking mindset meaning that he can sometimes get caught ‘offside’ in the rugby sense, i.e. closer to the opponents’ goal than his own, because he’s so keen to get forward, like a sort of unsynchronised gegen-pressing.
It’s apparent in his attacking play as well, how so often he’s in the inside-right channel, or on the edge of the six yard box. Fair enough, he gets the odd goal or assist from there, but as I think the stats above prove not often enough to justify how often he is in uber-advanced positions.
Here is in the 5-5 draw with Hibs in May 2018. Okay, Rangers score from this attack, but what is he doing in centre midfield? There’s nobody at all on the right hand side of the pitch!
I’ve touched on the ‘Tavernier only gets caught out of position defensively because he’s doing so much attacking’ trope; it’s a mechanism employed by people who are fans of his a fair bit I think to absolve Tavernier of his defensive responsibilities. Again, the number would appear to suggest that Rangers don’t lose a huge number of goals due to Tavernier’s wandering, but they did lose an important one in the first Old Firm derby of 2018-19 at Celtic Park.
This goal provides a fairly concise microcosm of things Tavernier does that drive me crazy. First, circled in red in the image above, he’s the second furthest forward of all Rangers’ players. He’s taken up a centre-forward position at the same time as his fellow on the left hand side, Borna Barisic, is even further forward. You can also see in the image that Rangers have thrown forward everyone that’s not a centre-back or a goalkeeper at this point in the match. Tavernier drifts infield, as does Scott Arfield, who was playing in behind the strikers. There was no-one covering the right side of the pitch at all at this point. That wouldn’t be an issue in this instance, but it does concern me that there’s a bit of miscommunication between the two players.
In a matter of seconds from now, Celtic clear the ball and break down the pitch, and the second element of Tavernier’s game that rips my knitting comes into play.
Celtic have a four on four break, and eventually score. Now, as the second-furthest player up the pitch, you could commend Tavernier for the tracking back he’s done. And the finish itself is probably due to Ovie Ejaria switching off and not tracking Ntcham. But why did Tavernier stop running when he did? You can see in the image above that he’s jogging back. Okay, it’s 80 odd yards he’s had to run, but Barisic and Ejaria have managed similar distances. And if he’s assessed the situation as four on four, then he’s still the wrong side of his man. Yes, Ejaria made a bigger mistake, but he’s a near 21-year old kid. Tavernier is six years older and the captain of the team. Should we not expect him to be trying a bit harder, especially in a derby against Celtic? It’s not uncommon for him to be jogging if he’s defending and sprinting if an attack is on the cards.
But this is what I mean about transitions. I suspect in the second image above, he’s waiting for the ball to be cleared out to him so he can instigate a counter-attack. I appreciate that a lot of these subjective conclusions exist where statistics can’t really help us out as much as we’d like them to, but for me Tavernier’s positioning travails and poor decision making do cost the club far more dearly than some would like to admit.
And I admit your mileage may vary on this, but from my point of view the following are are other examples of Tavenier causing consternation to his team-mates in big games.
Concentration: here he is in the Ufa match last season, oblivious to what’s going on around him.
Decision making: here, against Rapid, he chooses the highest risk option (trying to dribble between two defenders,) and loses possession.
In the Hearts match, he’s wandered so far out of position in transition, he may as well be in the changing room.
Tavernier’s hyper-attacking transitions close off Candeias’ options for a pass. Aberdeen win a penalty from the resultant turnover.
Glen Kamara took the blame for this hospital pass against Aberdeen, which led to the Dons scoring, but the ball was never less than 70:30 in Tavernier’s favour at any time. I feel he should have dealt with it.
In all of the scenarios above, Rangers conceded within a couple of minutes. This I feel is the high price of Tavernier’s roaming.
Conclusion
The tendency for Rangers fans to pick a player in the starting XI and single them out for abuse has frustrated me for many years because it seems so arbitrary and counter-productive, but I’m probably guilty of it myself. I felt a bit discomfited that I have identified my own scapegoat, to blame for the team’s defensive failings. While no doubt there’s a bit of cognitive dissonance going on, the reason I embarked on this piece was to prove to myself that there was some grain of objectivity in my frustration with Tavernier’s performances. To wit, these are my main bugbears;
- Doesn’t defend the back post
- Lets wingers get round him too easily
- Fast offensively, slow tracking back
- Lack of positional awareness
- Daydreaming
Does Tavernier’s offensive prowess outweigh his defensive lapses? I’m not convinced. By my reckoning he’s contributed to 34 open play goals in 3 seasons in the Scottish top flight, but has been involved in 21 defensive lapses directly leading to goals conceded in the same period. That’s a net gain, I suppose.
It’s always difficult to forecast what will happen over the course of an upcoming football season, but as we commence the second Steven Gerrard campaign, Rangers look to have a settled and parsimonious defence, combined with some exciting attacking talent. But what will this mean for Tavernier? Many of his fans seem to be expecting him to rack up the same number of goals and assists as in 2018-19, but as you might be able to tell, I think this is unlikely. Firstly, as he’s never really been an out-and-out goalscorer, Rangers would need to be awarded 14-15 penalties again. In terms of assists, Rangers appear to be moving away from a crossing-based approach, if their summer acquisitions are anything to go by. Jake Hastie, Jordan Jones, Greg Stewart, and Sheyi Ojo all tend to operate as inverted wingers, indicating that Rangers’ game plan for next season may be based on attempting to break defences down by the attacking three dribbling and shooting instead of relying on crossing as much, which didn’t appear to be as successful against Scottish Premiership sides last season.
We also at this stage don’t know what’s likely to happen with Alfredo Morelos, who has continually been linked with a move away from Ibrox. If he were to depart, or if Defoe is favoured as the centre forward, this also might impinge on Tavernier’s assist stats. The full-back’s partner-in-crime, Daniel Candeias has already left Rangers for Turkey. Many people, me included, feel the Portuguese took on a lot of Tavernier’s dirty work over the last two years, affording him a bit more of a free role. It’ll be interesting to see how a relationship with the replacement right midfielders/attackers develop.
There’s also the question about how much resource within the team Tavernier commits; in order to get your 17 goals and 20 assists he needs to take virtually all of the set-pieces, play in what is essentially a free role, and the team field a better than decent holding midfielder, right-sided mid, and inside-right to optimise his productivity. Is this sustainable for a club that desperately needs to win a league championship sooner rather than later?
A further consideration regarding the full-back is that as of writing he’s nearly 28; can he realistically improve his defensive game given deep down I’m not sure he wants to be a defender? If not a defender, could he play further up the pitch? I’m not convinced by that either; he more-or-less plays as a right winger at the moment.
So, is James Tavernier my scapegoat? Perhaps. However, analysis of statistics doesn’t preclude the existence of scapegoats entirely. Outliers in a dataset may tell us that something’s not quite right about an individual player’s performance, and I think there’s enough information out there to suggest that Tavernier’s value isn’t quite as high as it’s made out to be. 31 goal contributions in a season from a full-back does point towards regression, but I’m sure we’ll find out in time. Be sure to check back next summer for the third part of my whinge.
left_midfielder's Weblog 