2c:null 2d:["$","title",null,{"children":"Coca Cola Consultation Dataset | Dusty Wolfe Synek | Observable"}] 32:T691,{ const rows = stats.descriptives.filter(d => d.key !== "tenure"); const color = d => ({turnover: palette.leave, safety: palette.safety})[d.key] ?? palette.muted; const note = d => ({turnover: "low, so most are not planning to leave", conflict: "mild", safety: "fairly high"})[d.key]; return Plot.plot({ title: "On average, few plan to leave and teams feel safe", subtitle: "Average answer on each 7-point scale. The bar runs one standard deviation either side of the average.", width: 720, height: 230, marginLeft: 150, marginRight: 70, marginTop: 34, marginBottom: 44, style: {fontSize: "12px"}, x: {domain: [1, 7], ticks: 7, grid: true, label: "1 = strongly disagree · 7 = strongly agree", labelAnchor: "center", labelArrow: "none"}, y: {label: null, domain: rows.map(d => d.name), tickSize: 0}, marks: [ Plot.ruleX([4], {stroke: palette.ink, strokeOpacity: .5, strokeDasharray: "3,3"}), Plot.text([{x: 4}], {x: "x", text: () => "neutral", frameAnchor: "top", dy: -14, fill: palette.note}), Plot.ruleY(rows, {y: "name", x1: d => d.mean - d.sd, x2: d => d.mean + d.sd, stroke: color, strokeWidth: 5, strokeOpacity: .3, strokeLinecap: "round"}), Plot.dot(rows, {y: "name", x: "mean", r: 7, fill: color, stroke: "white", strokeWidth: 2, channels: {Mean: d => d.mean.toFixed(2), SD: d => d.sd.toFixed(2)}, tip: {format: {x: false, fill: false}}}), Plot.text(rows, {y: "name", x: "mean", text: d => d.mean.toFixed(2), dy: -17, fontWeight: 700, fontSize: 14, fill: palette.ink}), Plot.text(rows, {y: "name", x: d => d.mean + d.sd, text: note, dx: 12, textAnchor: "start", fill: palette.note}) ] }); } 33:T71a,{ const rows = d3.sort(stats.model.terms.slice(1), d => -Math.abs(d.beta)); const m = stats.model; const color = d => d.p < .05 ? palette.safety : palette.muted; const f = (v, n = 2) => v.toFixed(n).replace("-", "−").replace("0.", "."); const p = v => v < .001 ? "p < .001" : `p = ${f(v, 3)}`; return Plot.plot({ title: "Only psychological safety predicts intention to leave", subtitle: `Standardized effects with 95% confidence intervals. R² = ${f(m.r2, 3)}, adjusted R² = ${f(m.adj, 3)}, F(${m.k}, ${m.df}) = ${m.F.toFixed(2)}, ${p(m.pF)}.`, width: 720, height: 250, marginLeft: 150, marginRight: 170, marginTop: 34, marginBottom: 44, style: {fontSize: "12px"}, x: {domain: [-.8, .4], grid: true, label: "← lowers intention to leave raises it →", labelAnchor: "center", labelArrow: "none"}, y: {label: null, domain: rows.map(d => d.name), tickSize: 0}, marks: [ Plot.ruleX([0], {stroke: palette.ink}), Plot.text([{x: 0}], {x: "x", text: () => "no effect", frameAnchor: "top", dy: -14, fill: palette.note}), Plot.ruleY(rows, {y: "name", x1: "betaLo", x2: "betaHi", stroke: color, strokeWidth: 5, strokeOpacity: .45, strokeLinecap: "round"}), Plot.dot(rows, {y: "name", x: "beta", r: 7, fill: color, stroke: "white", strokeWidth: 2, channels: {"β": d => f(d.beta, 3), "95% CI": d => `${f(d.betaLo, 3)} to ${f(d.betaHi, 3)}`, p: d => p(d.p)}, tip: {format: {x: false, fill: false}}}), Plot.text(rows, {y: "name", x: () => .4, dx: 14, dy: -8, textAnchor: "start", fontWeight: 700, fill: palette.ink, text: d => `β = ${f(d.beta)}, ${p(d.p)}`}), Plot.text(rows, {y: "name", x: () => .4, dx: 14, dy: 8, textAnchor: "start", fill: palette.note, text: d => d.p < .05 ? "significant" : "not significant"}) ] }); } 34:Taec,{ const f = (v, n = 2) => v.toFixed(n).replace("-", "−").replace("0.", "."); const slope = Math.abs(stats.model.terms.find(d => d.key === "safety").b); const c = stats.correlations.find(d => d.key === "safety"); const leaning = d => d.turnover > 4; const count = v => `${v.filter(leaning).length} of ${v.length} lean toward leaving`; const low = D.filter(d => d.safety < 4.5), high = D.filter(d => d.safety >= 5.5); const labels = [`Leaning toward leaving (${D.filter(leaning).length})`, `Not leaning toward leaving (${D.length - D.filter(leaning).length})`]; const pts = d3.rollups(D, v => ({n: v.length, who: v.map(d => d.id).join(", ")}), d => d.safety, d => d.turnover) .flatMap(([x, ys]) => ys.map(([y, o]) => ({x, y, ...o}))); return Plot.plot({ title: "As safety falls, intention to leave climbs", subtitle: `Each one-point drop in safety goes with a ${slope.toFixed(2)}-point rise in intention to leave (r = ${f(c.r)}, ${c.p < .001 ? "p < .001" : "p = " + f(c.p, 3)}).`, caption: "Dot size shows how many employees share the same pair of scores.", width: 720, height: 460, marginBottom: 44, inset: 14, style: {fontSize: "12px"}, color: {legend: true, domain: labels, range: [palette.leave, palette.muted]}, r: {range: [4.5, 10]}, x: {domain: [1, 7], ticks: 7, grid: true, label: "Psychological safety (1 = low, 7 = high)", labelAnchor: "center", labelArrow: "none"}, y: {domain: [1, 7], ticks: 7, grid: true, label: "↑ Intention to turnover"}, marks: [ Plot.rect([{x1: 1, x2: 4.5, y1: 1, y2: 7}], {x1: "x1", x2: "x2", y1: "y1", y2: "y2", fill: palette.leave, fillOpacity: .07}), Plot.rect([{x1: 5.5, x2: 7, y1: 1, y2: 7}], {x1: "x1", x2: "x2", y1: "y1", y2: "y2", fill: palette.safety, fillOpacity: .06}), Plot.text([{x: 1.1, y: 1.15}], {x: "x", y: "y", textAnchor: "start", lineAnchor: "bottom", lineHeight: 1.3, fontWeight: 600, fill: palette.ink, text: () => `Low safety (below 4.5)\n${count(low)}`}), Plot.text([{x: 6.92, y: 6.88}], {x: "x", y: "y", textAnchor: "end", lineAnchor: "top", lineHeight: 1.3, fontWeight: 600, fill: palette.ink, text: () => `High safety (5.5 and up)\n${count(high)}`}), Plot.ruleY([4], {stroke: palette.ink, strokeOpacity: .5, strokeDasharray: "3,3"}), Plot.text([{x: 1.08, y: 4}], {x: "x", y: "y", text: () => "neutral", textAnchor: "start", dy: -8, fill: palette.note}), Plot.linearRegressionY(D, {x: "safety", y: "turnover", stroke: palette.ink, strokeWidth: 2, fill: palette.ink, fillOpacity: .08}), Plot.dot(pts, {x: "x", y: "y", r: "n", fill: d => labels[d.y > 4 ? 0 : 1], fillOpacity: .85, stroke: "white", strokeWidth: 1.5, channels: {Employees: "n", Who: "who"}, tip: {format: {fill: false, r: false}}}) ] }); } 35:T5df,{ const f = (v, n = 2) => v.toFixed(n).replace("-", "−").replace("0.", "."); const c = stats.correlations.find(d => d.key === "conflict"); const pts = d3.rollups(D, v => ({n: v.length, who: v.map(d => d.id).join(", ")}), d => d.conflict, d => d.turnover) .flatMap(([x, ys]) => ys.map(([y, o]) => ({x, y, ...o}))); return Plot.plot({ title: "Team conflict does not predict intention to leave", subtitle: `No relationship (r = ${f(c.r)}, p = ${f(c.p, 3)}). The fitted line is flat.`, caption: "Dot size shows how many employees share the same pair of scores.", width: 720, height: 320, marginBottom: 44, inset: 14, style: {fontSize: "12px"}, r: {range: [4, 9]}, x: {grid: true, label: "Intragroup conflict (1 = low, 7 = high)", labelAnchor: "center", labelArrow: "none", domain: [1, 7], ticks: 7}, y: {domain: [1, 7], ticks: 7, grid: true, label: "↑ Intention to turnover"}, marks: [ Plot.ruleY([4], {stroke: palette.ink, strokeOpacity: .5, strokeDasharray: "3,3"}), Plot.text([{y: 4}], {y: "y", text: () => "neutral", frameAnchor: "right", textAnchor: "end", dx: -4, dy: -8, fill: palette.note}), Plot.linearRegressionY(D, {x: "conflict", y: "turnover", stroke: palette.ink, strokeWidth: 2, fill: palette.ink, fillOpacity: .08}), Plot.dot(pts, {x: "x", y: "y", r: "n", fill: palette.muted, fillOpacity: .7, stroke: "white", strokeWidth: 1.5, channels: {Employees: "n", Who: "who"}, tip: {format: {r: false}}}) ] }); } 36:T5b5,{ const f = (v, n = 2) => v.toFixed(n).replace("-", "−").replace("0.", "."); const c = stats.correlations.find(d => d.key === "tenure"); const pts = d3.rollups(D, v => ({n: v.length, who: v.map(d => d.id).join(", ")}), d => d.tenure, d => d.turnover) .flatMap(([x, ys]) => ys.map(([y, o]) => ({x, y, ...o}))); return Plot.plot({ title: "Years with the company do not predict intention to leave", subtitle: `No relationship (r = ${f(c.r)}, p = ${f(c.p, 3)}). The fitted line is flat.`, caption: "Dot size shows how many employees share the same pair of scores.", width: 720, height: 320, marginBottom: 44, inset: 14, style: {fontSize: "12px"}, r: {range: [4, 9]}, x: {grid: true, label: "Years with the company", labelAnchor: "center", labelArrow: "none"}, y: {domain: [1, 7], ticks: 7, grid: true, label: "↑ Intention to turnover"}, marks: [ Plot.ruleY([4], {stroke: palette.ink, strokeOpacity: .5, strokeDasharray: "3,3"}), Plot.text([{y: 4}], {y: "y", text: () => "neutral", frameAnchor: "right", textAnchor: "end", dx: -4, dy: -8, fill: palette.note}), Plot.linearRegressionY(D, {x: "tenure", y: "turnover", stroke: palette.ink, strokeWidth: 2, fill: palette.ink, fillOpacity: .08}), Plot.dot(pts, {x: "x", y: "y", r: "n", fill: palette.muted, fillOpacity: .7, stroke: "white", strokeWidth: 1.5, channels: {Employees: "n", Who: "who"}, tip: {format: {r: false}}}) ] }); } 37:T64d,{ const order = ["Low safety (below 4.5)", "Moderate (4.5 to 5.25)", "High safety (5.5 and up)"]; const band = d => order[d.safety < 4.5 ? 0 : d.safety < 5.5 ? 1 : 2]; const rows = order.map(name => { const v = D.filter(d => band(d) === name), leaning = v.filter(d => d.turnover > 4).length; return {name, n: v.length, leaning, share: leaning / v.length, mean: d3.mean(v, d => d.turnover)}; }); const all = D.filter(d => d.turnover > 4).length; return Plot.plot({ title: `${Math.round(100 * rows[0].share)}% of low-safety employees lean toward leaving`, subtitle: `Share of each group scoring above neutral (4) on intention to turnover. Across the company it is ${all} of ${D.length}.`, width: 720, height: 340, marginTop: 56, marginBottom: 40, style: {fontSize: "12px"}, x: {label: null, domain: order, padding: .45, tickSize: 0}, y: {label: null, percent: true, tickFormat: d => d + "%", ticks: 4, grid: true, domain: [0, 60]}, marks: [ Plot.barY(rows, {x: "name", y: "share", rx: 4, fill: d => d.name === order[0] ? palette.leave : palette.muted, channels: {Employees: "n", "Average turnover intention": d => d.mean.toFixed(2)}, tip: {format: {fill: false, x: false}}}), Plot.text(rows, {x: "name", y: "share", text: d => Math.round(100 * d.share) + "%", lineAnchor: "bottom", dy: -22, fontSize: 22, fontWeight: 700, fill: palette.ink}), Plot.text(rows, {x: "name", y: "share", text: d => `${d.leaning} of ${d.n} employees`, lineAnchor: "bottom", dy: -6, fill: palette.note}), Plot.ruleY([0], {stroke: palette.ink}) ] }); } 38:T8d6,## Why this happens **Psychological safety.** Psychological safety is a shared belief that the team is safe for interpersonal risk-taking, such as admitting a mistake, asking for help or raising a problem (Edmondson, 1999). Our survey used four of the seven items in Edmondson's original scale. - Google studied 180 of its own teams and ranked psychological safety first of five team dynamics. People on teams with a stronger culture were also less likely to leave (re:Work Editorial Team, 2015). - In a Boston Consulting Group survey of 28,000 professionals in 16 countries, 12% of employees with the lowest safety said they were likely to quit within a year, against 3% where safety was high (Yousif et al., 2024). Our gap is wider, at 48% against 4%. - In Edmondson's field study, members of a low-safety team kept concerns to themselves, problems went unaddressed, and some coped by planning to leave the team (Edmondson, 1999). - Our average (5.30, SD 1.07) is close to what Edmondson found across 51 teams in one manufacturer (5.25, SD 1.03, on her full seven-item scale), so the company is typical on average. **Leadership and employee voice.** - Team leader coaching and organizational support both predicted a team's psychological safety (Edmondson, 1999). BCG names empathetic leadership as the way to build it (Yousif et al., 2024). - However open a manager believes they are, many employees are more likely to stay quiet than to question a decision or suggest an idea (Detert & Burris, 2016). - 52% of employees who left voluntarily said their manager or organization could have done something to keep them. 51% said no leader had talked with them about their job satisfaction or future in the three months before they left (McFeely & Wigert, 2019). **Turnover and its cost.** Replacing one employee costs one-half to two times that person's annual salary (McFeely & Wigert, 2019). In our data 17 of 100 employees score above neutral on intention to leave, and 11 of those 17 are in the low-safety group. **Why conflict is left out.** Conflict levels are mild (3.46) and unrelated to turnover intention. BCG lists "you have to avoid conflict" as a myth and treats unwillingness to dissent as a sign that a team is unsafe (Yousif et al., 2024). 39:T1314,## Glossary Every statistical term used in this notebook, with our own numbers as the example. ### Describing the data | Term | What it means | In our data | |---|---|---| | **n** | The number of people in the sample | 100 employees | | **Likert scale** | The 1 to 7 agree/disagree answer format | 1 = strongly disagree, 4 = neutral, 7 = strongly agree | | **Scale score** | One person's average across the questions in a scale | Safety is the average of 4 questions | | **Reverse-coded item** | A negatively worded question flipped so that a high number always means the same thing | Safety2R and Safety4R, already flipped in the file | | **Mean (M)** | The average | Turnover 2.84, conflict 3.46, safety 5.30, tenure 8.91 years | | **Standard deviation (SD)** | How spread out people are around the average. About two-thirds of people fall within one SD of the mean | Safety SD = 1.07, so most people score between about 4.2 and 6.4 | | **Cronbach's alpha (α)** | Whether the questions in a scale measure the same thing. It runs from 0 to 1, and .70 or higher counts as reliable | Turnover .92, safety .87, conflict .82 | ### Relationships between two variables | Term | What it means | In our data | |---|---|---| | **Correlation (r)** | The strength and direction of a straight-line relationship, from −1 to +1. Zero means no relationship | Safety and turnover: r = −.58 | | **Negative relationship** | As one goes up, the other goes down | Higher safety goes with lower intention to leave | ### The regression | Term | What it means | In our data | |---|---|---| | **Dependent variable** | The outcome we are trying to explain | Intention to turnover | | **Predictors** | The things we think explain it. Also called independent variables | Tenure, conflict, safety | | **Multiple regression** | A model that tests all predictors at once, so each effect is measured with the others held constant | One model with three predictors | | **R²** | The share of the differences in the outcome that the model explains | .343, about a third | | **Adjusted R²** | R² corrected for the number of predictors. It is always a little lower | .322 | | **F statistic** | Tests whether the model as a whole explains more than chance would | F(3, 96) = 16.70 | | **Degrees of freedom (df)** | The two numbers in the F brackets: the number of predictors, then n minus predictors minus 1 | 3 and 96 | | **B** | The unstandardized coefficient: how much the outcome changes for a one-point rise in the predictor | Safety B = −0.827. One more point of safety goes with 0.83 less intention to leave | | **β (beta)** | The standardized coefficient: the same effect in standard-deviation units, so predictors on different scales can be compared | Safety −.58, tenure −.05, conflict .04 | | **Constant** | The predicted outcome when every predictor is zero. Also called the intercept, and rarely interpreted | 7.14 | | **Standard error** | How uncertain a coefficient estimate is | Safety 0.118 | | **t** | The coefficient divided by its standard error. Further from zero is stronger evidence | Safety t = −6.99 | ### Deciding what is real | Term | What it means | In our data | |---|---|---| | **p-value** | The chance of seeing a result this strong if there were no real relationship | Safety p < .001 | | **Statistically significant** | p below .05, the usual cut-off for "unlikely to be chance" | Safety is significant. Tenure (p = .569) and conflict (p = .659) are not | | **95% confidence interval** | The range the true value plausibly falls in. If the range includes zero, the effect is not significant | Safety B runs from −1.06 to −0.59, which excludes zero | | **Cross-sectional** | A survey taken once, at one point in time. It can show that two things go together but cannot prove that one causes the other | Our survey | ### What we measured | Term | What it means | |---|---| | **Psychological safety** | A shared belief that the team is safe for interpersonal risk-taking, such as admitting mistakes, asking for help and raising problems (Edmondson, 1999) | | **Intention to turnover** | How much someone is thinking about or planning to leave. It measures intent and does not record who resigned | | **Intragroup conflict** | How much disagreement and friction a team has | | **Organizational tenure** | Years with the company | | **Leaning toward leaving** | Our own label for anyone scoring above neutral (4) on intention to turnover. 17 of 100 employees | | **Low-safety group** | Our own label for anyone scoring below 4.5 on psychological safety. 23 of 100 employees | ### Three numbers to remember - R² = .343: the model explains about a third of the differences in intention to leave. - Safety B = −0.83, p < .001: safety is the only predictor that matters. - 48% of low-safety employees lean toward leaving, against 4% of high-safety employees. 3a:T5cd,## References Baskin, K. (2023, June 14). *Four steps to building the psychological safety that high-performing teams need*. Harvard Business School Working Knowledge. https://www.library.hbs.edu/working-knowledge/four-steps-to-build-the-psychological-safety-that-high-performing-teams-need-today Delizonna, L. (2017, August 24). High-performing teams need psychological safety. Here's how to create it. *Harvard Business Review*. https://hbr.org/2017/08/high-performing-teams-need-psychological-safety-heres-how-to-create-it Detert, J. R., & Burris, E. R. (2016). Can your employees really speak freely? *Harvard Business Review, 94*(1), 80–87. https://hbr.org/2016/01/can-your-employees-really-speak-freely Edmondson, A. (1999). Psychological safety and learning behavior in work teams. *Administrative Science Quarterly, 44*(2), 350–383. https://doi.org/10.2307/2666999 McFeely, S., & Wigert, B. (2019, March 13). *This fixable problem costs U.S. businesses $1 trillion*. Gallup. https://www.gallup.com/workplace/247391/fixable-problem-costs-businesses-trillion.aspx re:Work Editorial Team. (2015). *Understand team effectiveness*. Google re:Work. https://rework.withgoogle.com/intl/en/guides/understand-team-effectiveness Yousif, N., Dartnell, A., May, G., & Knarr, E. (2024, January 4). *Psychological safety levels the playing field for employees*. Boston Consulting Group. https://www.bcg.com/publications/2024/psychological-safety-levels-playing-field-for-employees 3b:T121a,stats = { const ITEMS = {turnover: ["Turn1", "Turn2", "Turn3"], conflict: ["Conflict1", "Conflict2", "Conflict3"], safety: ["Safety1", "Safety2R", "Safety3", "Safety4R"]}; const NAMES = {turnover: "Intention to turnover", tenure: "Organizational tenure", conflict: "Intragroup conflict", safety: "Psychological safety"}; const col = k => D.map(d => d[k]); // regularized incomplete beta (Lentz continued fraction) gives exact t and F p-values const lgamma = x => { let y = x, t = x + 5.5, s = 1.000000000190015; t -= (x + .5) * Math.log(t); for (const c of [76.18009172947146, -86.50532032941677, 24.01409824083091, -1.231739572450155, .1208650973866179e-2, -.5395239384953e-5]) s += c / ++y; return -t + Math.log(2.5066282746310005 * s / x); }; const betacf = (x, a, b) => { const tiny = v => Math.abs(v) < 1e-30 ? 1e-30 : v; let c = 1, d = 1 / tiny(1 - (a + b) * x / (a + 1)), h = d; for (let m = 1; m <= 300; m++) { const m2 = 2 * m; for (const aa of [m * (b - m) * x / ((a - 1 + m2) * (a + m2)), -(a + m) * (a + b + m) * x / ((a + m2) * (a + 1 + m2))]) { d = 1 / tiny(1 + aa * d); c = tiny(1 + aa / c); h *= d * c; } if (Math.abs(d * c - 1) < 1e-13) break; } return h; }; const ibeta = (x, a, b) => { if (x <= 0) return 0; if (x >= 1) return 1; const bt = Math.exp(lgamma(a + b) - lgamma(a) - lgamma(b) + a * Math.log(x) + b * Math.log(1 - x)); return x < (a + 1) / (a + b + 2) ? bt * betacf(x, a, b) / a : 1 - bt * betacf(1 - x, b, a) / b; }; const pT = (t, df) => ibeta(df / (df + t * t), df / 2, .5); // two-tailed const pF = (F, d1, d2) => ibeta(d2 / (d2 + d1 * F), d2 / 2, d1 / 2); const tCrit = df => { let lo = 0, hi = 20; for (let i = 0; i < 60; i++) { const m = (lo + hi) / 2; pT(m, df) > .05 ? lo = m : hi = m; } return lo; }; const pearson = (x, y) => { const mx = d3.mean(x), my = d3.mean(y), r = d3.sum(x, (v, i) => (v - mx) * (y[i] - my)) / ((x.length - 1) * d3.deviation(x) * d3.deviation(y)); return {r, p: pT(r * Math.sqrt((x.length - 2) / (1 - r * r)), x.length - 2)}; }; const alpha = items => { const k = items.length; return k / (k - 1) * (1 - d3.sum(items, i => d3.variance(raw, r => r[i])) / d3.variance(raw, r => d3.sum(items, i => r[i]))); }; const inverse = A => { const n = A.length, M = A.map((r, i) => [...r, ...r.map((_, j) => +(i === j))]); for (let i = 0; i < n; i++) { let p = i; for (let r = i + 1; r < n; r++) if (Math.abs(M[r][i]) > Math.abs(M[p][i])) p = r; [M[i], M[p]] = [M[p], M[i]]; const v = M[i][i]; M[i] = M[i].map(x => x / v); for (let r = 0; r < n; r++) if (r !== i) { const f = M[r][i]; M[r] = M[r].map((x, j) => x - f * M[i][j]); } } return M.map(r => r.slice(n)); }; const ols = (yKey, xKeys) => { const y = col(yKey), X = D.map(d => [1, ...xKeys.map(k => d[k])]), n = y.length, k = xKeys.length, df = n - k - 1, p1 = k + 1; const XtXi = inverse(d3.range(p1).map(i => d3.range(p1).map(j => d3.sum(X, r => r[i] * r[j])))); const Xty = d3.range(p1).map(i => d3.sum(X, (r, q) => r[i] * y[q])); const b = XtXi.map(r => d3.sum(r, (v, j) => v * Xty[j])); const sse = d3.sum(X, (r, q) => (y[q] - d3.sum(r, (v, j) => v * b[j])) ** 2), sst = d3.variance(y) * (n - 1); const r2 = 1 - sse / sst, F = (r2 / k) / ((1 - r2) / df), tc = tCrit(df), sy = d3.deviation(y); return {n, k, df, r2, adj: 1 - (1 - r2) * (n - 1) / df, F, pF: pF(F, k, df), terms: ["(Constant)", ...xKeys].map((key, i) => { const se = Math.sqrt(sse / df * XtXi[i][i]), t = b[i] / se, s = i ? d3.deviation(col(key)) / sy : NaN; return {key, name: NAMES[key] ?? key, b: b[i], se, t, p: pT(t, df), lo: b[i] - tc * se, hi: b[i] + tc * se, beta: b[i] * s, betaLo: (b[i] - tc * se) * s, betaHi: (b[i] + tc * se) * s}; })}; }; const model = ols("turnover", ["tenure", "conflict", "safety"]); // check against an independent Python (statsmodels) run of the same model if (Math.abs(model.r2 - .3429) > 5e-4 || Math.abs(model.F - 16.70) > .01 || Math.abs(model.terms[3].b + .8270) > 5e-4) throw new Error("Regression does not match the statsmodels reference"); return { model, descriptives: ["turnover", "tenure", "conflict", "safety"].map(k => ({ key: k, name: NAMES[k], items: ITEMS[k]?.length ?? 1, mean: d3.mean(col(k)), sd: d3.deviation(col(k)), min: d3.min(col(k)), max: d3.max(col(k)), alpha: ITEMS[k] ? alpha(ITEMS[k]) : null })), correlations: ["tenure", "conflict", "safety"].map(k => ({key: k, name: NAMES[k], ...pearson(col(k), col("turnover"))})) }; } 3c:[{"id":"72f764f921576eb4460f059362e808108662dd5de0713cedd534d27056f4512e414a0875f35749cccba89de78d5f4e72940e5fce8c483cfec19368b7cda5ac54","href":"https://static.observableusercontent.com/files/72f764f921576eb4460f059362e808108662dd5de0713cedd534d27056f4512e414a0875f35749cccba89de78d5f4e72940e5fce8c483cfec19368b7cda5ac54","name":"coca-cola.csv","type":"text/csv","lastModified":1791168257656,"size":2930}] 30:{"id":"88010b4cd927fd63","slug":"coca-cola-consultation-dataset","title":"Coca Cola Consultation Dataset","thumbnail":"49e3b8e7b5c25c76853ac31b4ed41e895e2facb91573182f4e68e995d0ccb899","access_level":"public","create_time":"$D2026-10-05T02:41:41.782Z","update_time":"$D2026-10-05T16:38:56.758Z","view_time":null,"first_public_time":"$D2026-10-05T14:58:41.846Z","has_unpublished_edits":null,"fork":false,"stars":0,"starred":false,"owner":{"type":"team","id":"dff9b5e6e979adb0","login":"dlw272","name":"Dusty Wolfe Synek","avatar_url":null,"tier":"starter_2024"},"version":168,"publish_version":168,"role":null,"authored":false,"authors":[{"type":"individual","id":"318cf45c19986f2e","login":"dlw272","name":"Dusty Wolfe Synek","avatar_url":null,"tier":null}],"members":[{"type":"team","id":"dff9b5e6e979adb0","login":"dlw272","name":"Dusty Wolfe Synek","avatar_url":null,"tier":"starter_2024","role":"editor"}],"comments":[],"custom_thumbnail":false,"trashed":false,"lapsed":null,"subscribed":null,"forks":0,"collection":null,"type":"notebook","body":{"titling":"auto","title":"Coca Cola Consultation Dataset","cells":[{"id":1,"value":"# Coca Cola Consultation Dataset","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":5,"value":"## People who don't feel safe speaking up are the ones planning to leave\n\n**BGEN 342 Team 7 · Coca-Cola survey · n = 100 employees**\n\nWe tested three possible drivers of turnover intention: years with the company, conflict within teams, and psychological safety. Only psychological safety matters. Each one-point drop on the 7-point safety scale goes with a 0.83-point rise in intention to leave.\n","pinned":true,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":6,"value":"## 1. On average the company looks healthy\n\nThe average employee disagrees that they plan to leave (2.84 out of 7), reports mild conflict (3.46), and rates team safety at 5.30. Average tenure is 8.91 years. Those averages hide a group at the low end of safety, and that group is the problem.\n\nEach scale score is the average of its survey items, where 1 = strongly disagree, 4 = neutral and 7 = strongly agree. All three scales are reliable (Cronbach's α above .80), so the items can be combined.\n","pinned":true,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":13,"value":"$32","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":14,"value":"Inputs.table(stats.descriptives.map(d => ({...d, alpha: d.alpha == null ? \"N/A\" : d.alpha.toFixed(2)})), {\n columns: [\"name\", \"items\", \"mean\", \"sd\", \"min\", \"max\", \"alpha\"],\n header: {name: \"Variable\", items: \"Items\", mean: \"Mean\", sd: \"SD\", min: \"Min\", max: \"Max\", alpha: \"Cronbach's α\"},\n format: {mean: d3.format(\".2f\"), sd: d3.format(\".2f\"), min: d3.format(\".2f\"), max: d3.format(\".2f\")},\n align: {alpha: \"right\"}\n})","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":15,"value":"## 2. Safety is the only predictor that matters\n\nTogether the three predictors explain about a third of the differences in turnover intention (R² = .343, adjusted R² = .322), and the model is statistically significant, F(3, 96) = 16.70, p < .001. Nearly all of that comes from psychological safety (B = −0.827, β = −.58, p < .001). Tenure (p = .569) and conflict (p = .659) have no detectable effect.\n\nIn plain language, intention to turnover is negatively related to psychological safety. Employees who feel safe taking risks, admitting mistakes and raising tough issues are much less likely to be thinking about leaving. How long someone has worked here and how much their team argues tell us nothing about whether they plan to leave.\n\nIn the chart, a bar that crosses zero means we cannot say the predictor has any effect.\n","pinned":true,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":16,"value":"$33","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":18,"value":"Inputs.table([stats.model], {\n columns: [\"r2\", \"adj\", \"F\", \"pF\"],\n header: {r2: \"R²\", adj: \"Adjusted R²\", F: `F (${stats.model.k}, ${stats.model.df})`, pF: \"Significance\"},\n format: {r2: d3.format(\".3f\"), adj: d3.format(\".3f\"), F: d3.format(\".2f\"), pF: p => p < .001 ? \"p < .001\" : \"p = \" + d3.format(\".3f\")(p)}\n})\n","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":19,"value":"Inputs.table(stats.model.terms, {\n columns: [\"name\", \"b\", \"se\", \"beta\", \"t\", \"p\", \"lo\", \"hi\"],\n header: {name: \"Predictor\", b: \"B\", se: \"Std. error\", beta: \"β\", t: \"t\", p: \"p\", lo: \"95% CI low\", hi: \"95% CI high\"},\n format: {b: d3.format(\".3f\"), se: d3.format(\".3f\"), beta: d => isNaN(d) ? \"\" : d3.format(\".3f\")(d), t: d3.format(\".2f\"), p: p => p < .001 ? \"< .001\" : d3.format(\".3f\")(p), lo: d3.format(\".3f\"), hi: d3.format(\".3f\")}\n})\n","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":20,"value":"## 3. As safety falls, intention to leave climbs\n\nEach dot is one employee, or several with identical scores. The line is the best fit, with a 95% confidence band (r = −.58, p < .001).\n\nThe same chart for the other two predictors is flat: conflict (r = .06, p = .548) and tenure (r = −.07, p = .493). New and long-serving employees look alike, and so do people in high-conflict and low-conflict teams.\n","pinned":true,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":21,"value":"$34","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":22,"value":"$35","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":23,"value":"$36","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":24,"value":"","pinned":true,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":25,"value":"$37","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":26,"value":"## The core issue\n\nAbout a quarter of employees say they cannot raise problems or ask for help on their team, and they are the ones thinking about leaving. The company-wide safety average of 5.30 is fine, so a company-wide culture programme would mostly reach people who don't need it. Find the low-safety teams and fix those.\n\n**Limits of this analysis**\n\n- One survey at one point in time shows that low safety and intention to leave go together. It cannot prove that one causes the other.\n- The survey measures intention to leave. It does not record who resigned.\n- Three pairs of adjacent rows are identical (D011/D012, D043/D044, D089/D090) and may be duplicate entries. Removing them leaves the result the same: safety B = −0.831, R² = .346.\n- We also tested whether conflict hurts more when safety is low, and whether tenure has a curved effect. Neither was significant.\n","pinned":true,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":27,"value":"$38","pinned":true,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":28,"value":"$39","pinned":true,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":29,"value":"$3a","pinned":true,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":30,"value":"## Appendix: data and calculations\n\nThe last three cells below load the survey data, compute each scale score, and run the statistics. The last cell sets the chart colors.\n","pinned":true,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":8,"value":"raw = FileAttachment(\"coca-cola.csv\").csv({typed: true})\n","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":9,"value":"D = raw.map(r => ({\n id: r.ID,\n tenure: r.Tenure,\n turnover: d3.mean([r.Turn1, r.Turn2, r.Turn3]),\n conflict: d3.mean([r.Conflict1, r.Conflict2, r.Conflict3]),\n safety: d3.mean([r.Safety1, r.Safety2R, r.Safety3, r.Safety4R])\n}))\n","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":10,"value":"$3b","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":11,"value":"palette = ({leave: \"#e8590c\", safety: \"#1971c2\", muted: \"#7b898d\", ink: \"#1f2a2e\", note: \"#5b6b70\"})\n","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"}],"files":"$W3c","stdlib":"2"}} 31:"$30:body"