From the xG Spreadsheet to a Standard of Analysis: The Story of the Woman Who Brought Data into Chinese Football
Câu trả lời cốt lõi: Phân tích bóng đá dựa trên dữ liệu dùng xG để tách quá trình khỏi kết quả, PPDA để đo cường độ pressing, và các chỉ số tự xây như kiểm soát nguy hiểm để đo chất lượng triển khai bóng. Cách tiếp cận này giúp phát hiện chênh lệch giữa nhận thức thị trường và thực lực thực tế của đội bóng. Sự kiện chính: - Năm 2017, tại Bắc Kinh, một nhà phân tích cá cược đã dùng bảng xG tự tính (Guangzhou Evergrande 1,2 so với Shanghai SIPG 2,3) để đặt kèo SIPG +0,5 và thu về 40.000 nhân dân tệ. - Tại World Cup 2018, dữ liệu PPDA trận bán kết Pháp - Bỉ cho thấy Bỉ chịu 12,5 đường chuyền trước khi pressing, Pháp chỉ 8,2; bài phân tích đạt 500.000 lượt đọc. - Mùa 2020, dữ liệu mười năm chỉ ra lợi thế sân nhà giảm 37% khi không có khán giả; mô hình thắng 12/15 kèo nhưng thua 4 kèo liên tiếp do không cập nhật tham số. - Tại Euro 2021, chỉ số kiểm soát nguy hiểm cho thấy đội tuyển Ý dẫn đầu châu Âu với 18,2 pha bóng vào vùng 25 mét cuối trên mỗi trăm pha kiểm soát. - Phương pháp luận nhấn mạnh tương quan không phải nhân quả và yêu cầu kiểm tra biến gây nhiễu như lịch thi đấu trước khi kết luận. Nguồn: Phân tích chuyên sâu nội bộ, công bố ngày 31 tháng 8 năm 2026 | Cross-checked: VuaBong.vn Hỏi đáp liên quan: Hỏi: xG là gì và dùng để làm gì trong phân tích bóng đá? Đáp: xG là bàn thắng kỳ vọng, gán xác suất ghi bàn cho mỗi cú dứt điểm dựa trên vị trí và loại tiếp xúc, giúp tách chất lượng quá trình khỏi kết quả cuối cùng. Hỏi: PPDA thấp có nghĩa là đội bóng pressing tốt hơn? Đáp: Không nhất thiết, PPDA thấp chỉ cho biết đội pressing sớm hơn, còn ý nghĩa chiến thuật phụ thuộc vào kế hoạch trận đấu như trường hợp Pháp chủ động nhường bóng tại World Cup 2018. Hỏi: Chỉ số kiểm soát nguy hiểm được tính như thế nào? Đáp: Chia sân theo vùng dọc, gán trọng số theo tương quan với xác suất tạo cơ hội, rồi chia tổng trọng số cho số pha kiểm soát để ra số pha nguy hiểm trên mỗi trăm pha kiểm soát.
In 2026, at a café in Beijing's Chaoyang district, I sat across from three men who worked as betting analysts. On the table lay a printed Excel file, clipped inside a clear plastic sleeve: Guangzhou Evergrande's xG was 1.2, Shanghai SIPG's was 2.3. I pushed the paper toward them and spoke in Mandarin still carrying a heavy German accent: "You can laugh at me, but laugh after you've read the third column." One of them picked it up, glanced for ten seconds, put it down, and asked: "Where did you get these numbers?" I said: "I calculated them. Match by match. For three years." The result that day was 2-2. I had backed SIPG +0.5 and collected forty thousand renminbi. But what I took home was not the money — it was the sense that a door had opened, and behind that door was an entire profession.
I recount this detail not to praise myself. I recount it because it is the starting point of a method — and that method, eight years later, remains the only thing I trust when I sit down before a match that has not yet been played. In those eight years, Chinese football has passed through two major shocks, one pandemic, three changes of ownership at leading clubs, and a financial purge that forced nearly every club's balance sheet to be rewritten from scratch. Some of the men who laughed at me in that café are still in the business; some have left. And among them, there is a man who now calls me every Monday to ask whether last week's data table is finished.

That is why I still sit here writing long analytical pieces while most of the sports media industry has moved to three-hundred-word briefs. Because a brief answers the question "what happened." But the question "why did it happen, and in what form will it happen next week" — that question requires a spreadsheet, a model, and a person stubborn enough to stay seated after everyone else has gone home.
Context: When data becomes a condition of a league's existence
To understand why this method matters more in China than anywhere else, it must be placed in specific context. The Chinese Super League is a competition with financial characteristics entirely distinct from the major European leagues. Between 2026 and 2026, capital inflow grew exponentially: a club would pay a transfer fee for a South American striker several times the combined broadcasting revenue of the entire league. When money is not the constraint, the only remaining constraint is information — and information about the true quality of a team, in that period, was almost entirely absent from the public domain.
That is the gap I stepped into. Not because I was better than anyone, but because I came from a football culture — the Bundesliga — where match data had long been published down to the level of individual phases of play. In Germany, when I worked in the sports department of Belgrade Television in 2026, we had to count passes by hand on slow-motion videotape. Thirty years later, in Beijing, I realised I was doing exactly the same task — but this time with Excel, with a computer, and with a betting market large enough to pay for accuracy.
One point about the operating context must be made clear. Between 2026 and 2026, the level of data transparency in the CSL was far below that of the European leagues. Basic metrics such as shot counts and possession were published, but information on shot locations, chance quality, and passes allowed before recovery — the things that give a model its actual value — was not. An analyst was forced either to build a private data collection system or to accept analysis on a defective foundation. I chose the first, and the price was time: for each match, I spent an average of four hours manually coding every meaningful phase of play, classified by zone, by situation, by pressure.
This may sound extreme. But look at the consequences. When a league has large capital flows but poor data, the betting market prices teams on sentiment and on star reputation. This is precisely the systematic gap. A team with three attacking stars but a loose midfield will always be priced above its true strength, because bookmakers and the public look at names. A team with no stars at all but good pressing organisation will always be underpriced. The difference between the two is where data creates advantage. That is not predicting football — it is measuring the gap between market perception and on-pitch reality, and betting that the gap will correct itself.
But there is one variable I have learned to respect: policy. The annual CSL season does not operate in a vacuum. Regulations on salary caps, on the number of foreign players, on requirements to field young players, on transfer spending limits — every change in these alters the structure of the analytical problem. A data table built under a three-foreign-player regime loses its value the moment the regime becomes five. An analyst who does not update parameters will lose, and lose systematically rather than randomly.
Core: PPDA, dangerous control, and the two metrics I built myself
At this point the tools must be stated plainly, because saying only "I use data" is not analysis, it is a slogan.
The first metric is xG — expected goals. The basic principle: every shot is assigned a probability of becoming a goal based on location, angle, type of contact (foot or head), and in fuller versions, based on defender pressure and goalkeeper position. The sum of probabilities across all shots in a match is that team's xG. Its meaning lies in separating process from outcome: a team with xG of 2.3 but only one goal is playing better than the result shows; a team with xG of 0.4 but two goals is living on luck, and luck is something that tends to redistribute over time.
During the 2026 season, I built a standard table template for every CSL match with four foundation columns: home xG, away xG, the difference in shots inside the penalty area, and a possession index weighted by zone. The fourth column was my own invention, and I gave it a simple name: dangerous control. The method: divide the pitch into vertical zones, assign each zone a weight based on the statistical correlation between the number of phases entering that zone and the probability of creating a goalscoring chance, then divide the total weight by the number of possession phases. The result is a number showing how many genuinely dangerous phases a team generates per hundred possession phases.
The second metric, more important in modern football, is PPDA — passes allowed per defensive action. The lower the PPDA, the higher and earlier the team presses. This is the metric I use to end an eternal argument in Chinese football: which team has greater "fighting spirit." Spirit cannot be measured; PPDA can. And when measured, the result often contradicts perception.
Let me give a concrete example from my own tracking experience. In the summer of 2026, during the World Cup in Russia, I dissected the France-Belgium semi-final using PPDA data across the tournament. The result: Belgium allowed opponents 12.5 passes before pressing; France only 8.2. Conventional reading would say France pressed better and was more proactive. But a closer reading of the match shows the opposite: France deliberately conceded the ball, drew Belgium forward, and counter-attacked at maximum speed. France's low PPDA was not an expression of aggression — it was an expression of a plan. I wrote an article with that argument, it was shared by a European magazine, and it reached half a million reads. The match ended 1-0 to France from a set piece. The argument was not wrong, but I must be honest: the goal came from a phase my model did not predict, and that is a limitation that belongs in the closing section of an article.
With Chinese football, PPDA has another application: it is an instrument for detecting physical decline. A team pressing high in round 5 and still pressing high in round 25 is a team with correctly built physical foundations. A team whose PPDA rises round by round — meaning its pressing grows looser — is paying the price for a congested fixture list or poor rotation. This metric does not judge the coach; it merely records the truth that players' bodies have limits, and that limit appears in the data table before it appears on the scoreboard.
One further metric deserves mention — one I applied to the Italian national team at Euro 2026, because it illustrates how I extend the analytical frame. Mancini's Italy held around 60 percent possession but public opinion called the approach harmless. I applied the very concept of dangerous control I had built for the CSL and obtained the result: Italy led Europe with 18.2 phases entering the final 25 metres per hundred possession phases. That number explains why Italy's possession was not as harmless as it felt. I backed Italy to win at 11/1 and won. But more important than the money was the fact that a metric built for one specific league — the CSL — could transfer to an entirely different league and retain meaning, provided its definition was clear enough. That is the standard I set for every metric I use: if I cannot explain the definition to someone who has never heard of it, the metric is not good enough.
Contrarian angle: correlation is not causation, and that is why most football analysis is wrong
Now the section I most want to state plainly, because it is the section this industry rarely agrees to state.
In the 2026 season, the pandemic froze global football. My data supply contract was cut by sixty percent. During that period I built a predictive model from ten years of historical data, focused on one variable: home advantage. When the Bundesliga returned in May 2026 with matches behind closed doors, the data showed home advantage had fallen by thirty-seven percent. I bet according to that model and won twelve of fifteen wagers.
But then I lost four consecutive wagers. The cause was not that the model was wrong, but that I was stubborn: I refused to update parameters after the first three rounds, believing the sample was still too small to adjust. That was a methodological error. In a fast-changing system, clinging to a large sample can be a form of procrastination — and the market does not procrastinate with you.
That lesson has shaped how I write ever since. At the end of every analytical piece, I add a section with a fixed name: assumptions and latency. That section lists what the model assumes, what it does not yet account for, and which parameters would have to change if new data appeared. I keep the original analytical frame, but add an update mechanism according to a predefined procedure rather than according to inspiration.
This is also why I do not accept conclusions based on correlation alone. A team winning many matches when player X starts does not prove player X is the cause. It may be that X only started in easy matches, or that X's inclusion coincided with a favourable fixture run. An analyst has a duty to check for confounding variables before speaking. In the CSL context, the most common confounder is the fixture list: a team can look very strong across a run of matches in which three consecutive opponents were all in a period of squad crisis. Looking at that winning run and concluding something about tactical strength is a systematic error.
And here is the largest contrarian point, the one I know will irritate some readers: most football arguments are not arguments about football. They are arguments about who has the right to make judgements. When a coach is criticised for "having no tactics," the critic usually cannot produce a verifiable alternative description. When a player is called "lacking spirit," the person making the call usually cannot cite any metric that measures spirit. I am not saying those judgements are necessarily wrong — I am saying they are unverified, and in an industry where every decision carries economic value, unverified means not yet qualified to conclude.
I must also be honest about the limits of my own method. xG cannot measure psychology. PPDA cannot measure decision quality. Dangerous control cannot distinguish a dangerous phase created by good attacking organisation from one created by an individual error in the opposing defence. Data describes structure, not people. An analyst who uses data and forgets that will produce conclusions that are statistically precise but footballing meaningless. I have made that mistake, and the correction is to always set data alongside video review — not to verify the numbers, but to learn what the numbers are leaving out.
One further variable I am obliged to mention, though it does not appear in the spreadsheet: the position of women in this industry. In 2026, when I presented the xG table in that café, the first reaction was not technical rebuttal but a question about qualifications. I did not argue with that question. I handed over the spreadsheet. And the spreadsheet, over time, answered more effectively than any argument I could have spoken aloud. But I will not pretend it did not happen, because if the next generation still faces the same question, then my silence would be a small betrayal of the profession I chose.
Takeaway: signals of the next round
There is one sentence I still say to my team of three associates each time we finish the round's data table: do not ask who will win this match, ask what this match is being mispriced about. The first question is a fan's question. The second is an analyst's question, and it always has a concrete answer if you are willing to spend the time.
In an annual season, the most important signal is not the result of the round just played, but the movement of process metrics across three consecutive rounds. A team holding its PPDA steady while the number of matches rises is a team with genuine physical foundations. A team whose dangerous control declines while still winning is a team consuming luck — and luck, by the definition of probability, is an asset with an expiry date.
The question I leave with the reader, and the question I ask myself before every round: when you watch a match and feel you understand it, are you understanding the match, or are you understanding the story you had already prepared in your head? Data will not answer that question for you. But it is the only thing that makes the question answerable.
