Skip to content
ContentLora

    Tip: press / anywhere to search.

    Explainer

    Statistics and evidence: a crash course

    A poll or survey interviews a sample chosen to represent a larger population, and only probability samples let pollsters calculate a margin of sampling error.[1][2] Many common mistakes come from misreading a number: a correlation does not automatically mean causation, a p-value does not measure the size or importance of an effect, and the mean is more sensitive to a few extreme values than the median.[3][4][5] Official statistics are governed by published principles on impartiality and confidentiality; many economic series are seasonally adjusted or adjusted for inflation, and early GDP estimates are updated as more source data arrive.[6][7][8][9][10]

    Editor reviewedUpdated Explainer63 claims · 100% primaryStatistics and evidence

    This crash course covers how to read the numbers in news stories, studies and official releases: where survey figures come from, how much to trust them, which summary statistics mislead, and how government agencies produce their data. Work through the course list in order, or jump to the section you need. The section on test results is general information, not medical advice: talk to a clinician about what a specific result means for you.

    Samples and polls

    Polls and surveys start with a sample: a subset of a population selected to represent the whole, whose answers are used to estimate characteristics of everyone.[1] In a probability sample, every unit has a chance of selection that can be calculated, which is what makes population estimates possible.[11] In a non-probability sample, such as a quota, convenience or volunteer sample, some people have no chance of selection, or their chance cannot be determined, so sampling error cannot be estimated.[12] Online opt-in polls fall into that second group: respondents are recruited from sources such as social media ads, reward websites or self-enrolment rather than chosen at random.[13]

    How a question is asked also matters: small wording differences can substantially change answers. In a January 2003 Pew survey, 68% favoured military action in Iraq, but only 43% did when the question added that US forces might suffer thousands of casualties.[14] Question order and whether a question is open or closed can also shift answers.[15][16] The American Association for Public Opinion Research (AAPOR) asks researchers who release results to disclose the sponsor, the exact wording, the population, how the sample was drawn, the dates, sample sizes and the weighting.[17] The polls explainer covers panels, address-based sampling and how pollsters’ methods have changed.

    Margin of error

    A survey result will probably not exactly match the true population value, and the margin of sampling error describes how close it can reasonably be expected to fall.[18] A margin of plus or minus 3 points at 95% confidence means that if the same survey were fielded 100 times, the result would be expected to fall within 3 points of the true value in 95 of them.[19] A simple random sample of 1,067 people gives about that margin.[20]

    Four points are often missed:

    • Leads have bigger margins. The margin for the difference between two candidates is generally about twice the margin reported for each one.[21]
    • Subgroups have bigger margins. In Pew’s example, a group of about 160 people inside a 1,067-person sample carries a margin of plus or minus 8 points.[22]
    • It applies only to probability samples. AAPOR says it does not apply to opt-in online polls or other non-probability polls.[2]
    • It is not the only error. Question wording, who is reached and likely-voter models can also introduce error, and these are hard to quantify and rarely reported.[23] AAPOR says there is no measurable overall margin of error for a poll.[24]

    Pew says trends across a number of polls provide more confidence than one or two polls.[25] The margin of error explainer works through the arithmetic.

    Weighting

    Polls without adjustment tend to over-represent people who are easier to reach, so pollsters weight their data to match the population, usually on several demographic measures.[26] Raking, which repeatedly adjusts each respondent’s weight until the sample matches known population distributions, is the most common method in public opinion surveys.[27] Weighting has a cost: it makes the margin of error larger, an increase statisticians call the design effect.[28]

    Weighting cannot fix everything. In Pew’s 2018 study of more than 30,000 online opt-in interviews, no weighting procedure brought average bias across 24 benchmarks below 6 percentage points, down from 8.4 points unweighted.[29] Choosing the right weighting variables mattered more than choosing the statistical method.[30] The weighting explainer compares raking, matching and propensity scores.

    Averages

    “The average” usually means the mean: the sum of all values divided by the number of values.[31] The median is the middle value when the data are sorted, and the mode is the most common value.[32][33] Because it includes every value, the mean is more sensitive to outliers, and the median is usually preferred when a distribution is not symmetrical.[5]

    Income is the classic case. The UK Office for National Statistics says a few very high incomes can influence the mean, while the median gives a good indication of the typical person’s standard of living.[34] The two can move differently: over the 10 years to the financial year ending 2024, UK median household disposable income grew 7.0% while mean income grew 0.6%.[35] The averages explainer shows which average fits which data.

    Correlation, causation and significance

    A correlation measures the size and direction of a relationship between variables, but it does not automatically mean that a change in one causes the change in the other.[36][3] Ice cream and sunscreen sales rise and fall together because of the season, not because one affects the other.[37] A controlled study is the most effective way to establish causality, and because ethical limits restrict controlled studies, observational studies are often used instead.[38][39] The correlation explainer has more examples, and the epidemiology explainer shows how health researchers weigh observational evidence.

    A p-value is the probability of getting a result at least as extreme as the one observed if the null hypothesis, typically no effect, is true.[40] Results with p below 0.05 are often called “statistically significant”, but the Cochrane Handbook says that threshold is arbitrary.[41] In 2016 the American Statistical Association issued six principles on p-values.[42] They include that a p-value does not measure the probability that the studied hypothesis is true, and that it does not measure the size of an effect or the importance of a result.[43][4] Cochrane also warns that a p-value above 0.05 is often misread as evidence of no effect, and tells authors to look at the effect estimate and its confidence interval instead.[44] See the p-values explainer, and the relative vs absolute risk explainer for how effect sizes are reported.

    Base rates and false positives

    A test’s sensitivity is the probability of a positive result when the person has the condition, and its specificity is the probability of a negative result when they do not.[45] A different question is answered by the positive predictive value: the probability that someone who tests positive actually has the condition.[46] That depends on the base rate. For a given test, the lower the prevalence in the group screened, the lower the positive predictive value.[47] The National Cancer Institute says common screening tests such as mammography and the PSA test have false-positive rates of 5% to 10% per screen, higher with repeat screening.[48] The base rates explainer works through an example, and the cancer screening explainer covers the tests themselves.

    Reading charts

    UK government guidance says the numerical axis of a bar chart should not start above zero, because readers perceive bars as proportional to each other.[49] It cites an HM Treasury chart whose axis started at 8%, which exaggerated a fall in inflation from 11.1% to 10.1%.[50] It also advises against dual-axis line charts, which it says can be manipulated to show almost any relationship between two series.[51] Its guidance on uncertainty warns that changes between periods can fall within survey error and random variation.[52] The chart guide has a checklist.

    Official statistics

    The UN Fundamental Principles of Official Statistics, endorsed by the UN General Assembly in 2014, say official statistics should be compiled and made available impartially and that individual data collected by statistical agencies are strictly confidential.[53][6][7] In the United States, a federal rule (5 CFR part 1321), which from December 2024 supersedes conflicting guidance, requires recognised federal statistical agencies to produce relevant and timely statistics, protect confidential data and publish release schedules in advance.[54][55][56] Early estimates can be updated: the Bureau of Economic Analysis estimates each quarter’s GDP three times, with the second and third estimates adding source data that were not available earlier.[10] The official statistics explainer covers the rules and the revision cycle.

    Two adjustments appear in most economic releases:

    • Seasonal adjustment removes regular calendar-related patterns, such as higher retail sales every December, so that consecutive periods can be compared.[57] The Census Bureau maintains X-13ARIMA-SEATS, software used for this.[58]
    • Real vs nominal. Nominal values are in current prices; real values are adjusted for inflation so they can be compared over time.[9] The headline GDP growth figure is the change in real GDP.[59] The CPI explainer shows how the most common price index is built.

    Developing in Science & Energy

    Questions readers ask

    What does "plus or minus 3 points" in a poll mean?

    It is the margin of sampling error. At the usual 95% confidence level, if the same survey were fielded 100 times, the result would be expected to fall within 3 percentage points of the true population value in 95 of them. It covers only sampling error, not problems such as question wording or who was reached.[19][23]

    Is a 5-point lead outside a 3-point margin of error a real lead?

    Not necessarily. The margin for the difference between two candidates is generally about twice the margin reported for each one. In Pew's example, a 5-point lead with a 3-point margin per candidate could plausibly be anywhere from -1 to +11 points.[21][60]

    Why are polls weighted?

    Raw samples tend to over-represent people who are easier to reach, and some groups are deliberately oversampled. Weighting aligns the sample with known population characteristics such as sex, age, race and ethnicity, education and region.[61][26][62]

    Does a correlation prove causation?

    Not automatically. The Australian Bureau of Statistics says a controlled study, in which comparable groups receive different treatments, is the most effective way to establish causality, and observational studies are often used instead when ethical limits rule such studies out.[3][38][39]

    Does "statistically significant" mean a result is important?

    No. The American Statistical Association says a p-value, or statistical significance, does not measure the size of an effect or the importance of a result. In a large study a small p-value may reflect a trivial effect.[4][63]

    When should I use the median rather than the mean?

    The median is usually preferred when a distribution is not symmetrical, because the mean is more sensitive to outliers. The UK Office for National Statistics says median income gives a good indication of the typical person's standard of living, while a few very high incomes can pull up the mean.[5][34]

    Why does a positive result from an accurate test not always mean the condition is present?

    Because of the base rate. For a given test, the lower the prevalence of a condition in the group tested, the lower the chance that a positive result is a true positive.[46][47]

    Sources

    · 63 claims from 14 publishers

    Each numbered claim is a statement we checked against the sources listed with it. Tap a number in the text to see its sources in place.

    1. [1]

      A sample is a subset of a population selected to represent the whole, and information from it is used to estimate characteristics of the entire population. confirmedas of 2026-11-14

      • Census and sample · Australian Bureau of Statistics · Primary source · Census and sample (retrieved 2026-11-14)
      • Census and sample · Australian Bureau of Statistics · Primary source · Census and sample (retrieved 2026-11-14)
    2. [2]

      AAPOR says the margin of sampling error applies only to probability-based surveys, in which participants have a known and non-zero chance of selection, and not to opt-in online surveys or other non-probability polls. confirmedas of 2026-11-14

    3. [3]

      A correlation between two variables does not automatically mean that a change in one causes the change in the other. confirmedas of 2026-11-14

    4. [4]

      The ASA's fifth and sixth principles are that a p-value, or statistical significance, does not measure the size of an effect or the importance of a result, and that a p-value by itself is not a good measure of evidence about a model or hypothesis. confirmedas of 2026-11-14

    5. [5]

      Because it includes every value, the mean is more sensitive to outliers than the median or mode, and the median is usually preferred when a distribution is not symmetrical. confirmedas of 2026-11-14

    6. [6]

      The first UN principle says official statistics that meet the test of practical utility are to be compiled and made available impartially, to honour citizens' entitlement to public information. confirmedas of 2026-11-14

    7. [7]

      The UN principles say individual data collected by statistical agencies are to be strictly confidential and used only for statistical purposes. confirmedas of 2026-11-14

    8. [8]

      The Census Bureau defines seasonal adjustment as estimating the seasonal component of a time series, and where relevant trading-day and moving-holiday effects, and removing them, so that movements are easier to analyse over consecutive periods. confirmedas of 2026-11-14

    9. [9]

      Nominal values are measured in current prices, while real values are adjusted for inflation by measuring them in constant prices, which lets them be compared across time. confirmedas of 2026-11-14

    10. [10]

      BEA estimates each quarter's GDP three times; the advance estimate comes about a month after the quarter ends, and the second and third estimates add source data that were not available earlier. confirmedas of 2026-11-14

    11. [11]

      In a probability (random) sample every unit in the population has a chance of selection that can be calculated, which makes it possible to produce population estimates. confirmedas of 2026-11-14

      • Census and sample · Australian Bureau of Statistics · Primary source · Random sampling (retrieved 2026-11-14)
      • Census and sample · Australian Bureau of Statistics · Primary source · Random sampling (retrieved 2026-11-14)
    12. [12]

      In a non-probability sample, such as a convenience, quota or volunteer sample, some people have no chance of selection or their chance cannot be determined, so sampling error cannot be estimated. confirmedas of 2026-11-14

      • Census and sample · Australian Bureau of Statistics · Primary source · Non-random sampling (retrieved 2026-11-14)
      • Census and sample · Australian Bureau of Statistics · Primary source · Non-random sampling (retrieved 2026-11-14)
    13. [13]

      Pew describes online opt-in polls as convenience samples whose respondents are not selected at random but recruited from online sources such as social media or search ads, reward websites or self-enrolment in a panel. confirmedas of 2026-11-14

    14. [14]

      Small differences in question wording can substantially change answers; in a January 2003 Pew survey, 68% favoured military action in Iraq, but only 43% did when the question added that US forces might suffer thousands of casualties. confirmedas of 2026-11-14

    15. [15]

      The order of questions can influence answers, because earlier questions can provide context for later ones; Pew randomises many lists of answer options for this reason. confirmedas of 2026-11-14

    16. [16]

      Whether a question is open-ended or closed-ended can change results; after the 2008 election, 58% chose the economy as their top issue when it was offered as an option, but only 35% volunteered it in the open-ended version. confirmedas of 2026-11-14

      • Writing Survey Questions · Pew Research Center · Primary source · Open- and closed-ended questions (retrieved 2026-11-14)
    17. [17]

      AAPOR's Code of Professional Ethics and Practice asks researchers who release results to disclose who sponsored and conducted the research, the exact question wording, the population, how the sample was drawn, the dates, sample sizes and how the data were weighted, and AAPOR says the Code applies to all public opinion researchers, members or not. confirmedas of 2026-11-14

      • Disclosure Standards · American Association for Public Opinion Research · Primary source · Section III, A (retrieved 2026-11-14)
      • Disclosure Standards · American Association for Public Opinion Research · Primary source · Section III, A (retrieved 2026-11-14)
      • Disclosure Standards · American Association for Public Opinion Research · Primary source · FAQ (retrieved 2026-11-14)
    18. [18]

      Pew Research Center explains that because a survey talks only to a sample, its result will probably not exactly match the true population value, and the margin of sampling error describes how close the result can reasonably be expected to fall to that value. confirmedas of 2026-11-14

    19. [19]

      A margin of error of plus or minus 3 percentage points at the 95% confidence level means that if the same survey were fielded 100 times, the result would be expected to fall within 3 points of the true population value in 95 of them. confirmedas of 2026-11-14

    20. [20]

      A simple random sample of 1,067 people has a margin of error of plus or minus 3 percentage points for estimates of overall support for a candidate. confirmedas of 2026-11-14

    21. [21]

      To judge whether one candidate's lead is real, the relevant margin of error is the one for the difference between the two candidates, which is generally about twice the margin reported for each candidate. confirmedas of 2026-11-14

    22. [22]

      Subgroups have larger margins of error than the full sample; Pew's example is a group of about 160 people within a 1,067-person sample, giving plus or minus 8 points for a single estimate and plus or minus 16 points for the gap between two candidates. confirmedas of 2026-11-14

    23. [23]

      Sampling error is only one source of survey error; question wording, who is reached, how candid respondents are and likely-voter models can also introduce error, and unlike sampling error these are hard to quantify and rarely reported. confirmedas of 2026-11-14

    24. [24]

      AAPOR says there is no measurable overall margin of error for a poll, because other errors such as question design and interviewing cannot be measured. confirmedas of 2026-11-14

    25. [26]

      Without adjustment, polls tend to over-represent people who are easier to reach and under-represent those who are harder to interview, so pollsters weight their data to match the population, usually on several demographic measures. confirmedas of 2026-11-14

    26. [27]

      Raking, or iterative proportional fitting, is the most common weighting method in public opinion surveys; it repeatedly adjusts each respondent's weight until the sample matches known population distributions on the chosen variables. confirmedas of 2026-11-14

    27. [28]

      Weighting helps avoid biased results but also makes the margin of error larger; statisticians call this increase in variability the design effect, and a margin of error that ignores it claims more precision than the survey warrants. confirmedas of 2026-11-14

    28. [29]

      In Pew's 2018 study of more than 30,000 online opt-in interviews, no weighting procedure brought average bias across 24 benchmarks below 6 percentage points, down from 8.4 points unweighted, so even the best approach removed only about 30% of the bias. confirmedas of 2026-11-14

    29. [30]

      Pew found that choosing the right weighting variables mattered more for accuracy than choosing the statistical method; adding political variables such as party identification cut more bias, while complex methods never did more than 0.3 points better than simple raking. confirmedas of 2026-11-14

    30. [31]

      The mean is the sum of all the values divided by the number of values; it is also called the arithmetic average and is the measure most commonly meant by "the average". confirmedas of 2026-11-14

    31. [32]

      The median is the middle value when the data are sorted, with half the values below it and half above; with an even number of values it is the mean of the two middle values. confirmedas of 2026-11-14

    32. [33]

      The mode is the most common value in a dataset; there can be more than one, and unlike the mean and median it can be used for categorical data. confirmedas of 2026-11-14

    33. [34]

      The UK Office for National Statistics says mean income can be influenced by a few individuals with very high incomes and so may not reflect the typical person's living standard, while the median, the middle person's income, gives a good indication of the typical individual's standard of living. confirmedas of 2026-11-14

    34. [35]

      Over the 10 years to the financial year ending 2024, UK median household disposable income grew 7.0% while mean income grew 0.6%. confirmedas of 2026-11-14

    35. [36]

      The Australian Bureau of Statistics defines correlation as a statistical measure, expressed as a number, of the size and direction of a relationship between two or more variables. confirmedas of 2026-11-14

    36. [37]

      The ABS gives the example of ice cream and sunscreen sales, which rise and fall together over the year because of the season, not because one affects the other. confirmedas of 2026-11-14

    37. [38]

      The ABS says a controlled study, in which a population is split into two groups comparable in almost every way that then receive different treatments, is the most effective way to establish causality. confirmedas of 2026-11-14

      • Correlation and causation · Australian Bureau of Statistics · Primary source · Establishing causation (retrieved 2026-11-14)
      • Correlation and causation · Australian Bureau of Statistics · Primary source · Establishing causation (retrieved 2026-11-14)
    38. [39]

      Ethical limits on controlled studies mean observational studies are often used instead to investigate correlation and causation, adding evidence to the wider process of deciding whether a causal link exists. confirmedas of 2026-11-14

      • Correlation and causation · Australian Bureau of Statistics · Primary source · Establishing causation (retrieved 2026-11-14)
      • Correlation and causation · Australian Bureau of Statistics · Primary source · Establishing causation (retrieved 2026-11-14)
    39. [40]

      A p-value is the probability of getting a test result at least as extreme as the one observed if the null hypothesis (typically, no effect or no difference) is true. confirmedas of 2026-11-14

    40. [41]

      P-values below 0.05 are often reported as "statistically significant", but the Cochrane Handbook says the 0.05 threshold is arbitrary and became common largely because test statistics used to be compared against printed tables. confirmedas of 2026-11-14

    41. [42]

      On 7 March 2016 the American Statistical Association released a statement on statistical significance and p-values setting out six principles for their proper use and interpretation; ASA president Jessica Utts said it was the first time the community of statisticians, as represented by the ASA Board, had issued a statement on these issues. confirmedas of 2026-11-14

    42. [43]

      The ASA's first two principles are that p-values can indicate how incompatible the data are with a specified statistical model, and that they do not measure the probability that the studied hypothesis is true or that the data were produced by random chance alone. confirmedas of 2026-11-14

    43. [44]

      The Cochrane Handbook warns that a p-value above 0.05 is often misread as evidence that an intervention has no effect, and tells authors to look at the effect estimate and its confidence interval instead. confirmedas of 2026-11-14

    44. [45]

      The National Cancer Institute defines a screening test's sensitivity as the probability of a positive result when the person has cancer, and its specificity as the probability of a negative result when the person does not. confirmedas of 2026-11-14

    45. [46]

      The positive predictive value (PPV) is the probability that a person who tests positive has the disease, and the negative predictive value (NPV) is the probability that a person who tests negative does not. confirmedas of 2026-11-14

    46. [47]

      Predictive values are linked to sensitivity and specificity by Bayes' rule, with the prevalence of the disease in the screened group as the starting (prior) probability; for a given test, the lower the prevalence, the lower the PPV and the higher the NPV. confirmedas of 2026-11-14

    47. [48]

      NCI says commonly used screening tests such as mammography and the PSA test have false-positive rates of 5% to 10% per screen, and that cumulative false-positive rates with repeat screening are substantially higher. confirmedas of 2026-11-14

    48. [49]

      UK government guidance says the numerical axis of a bar chart should not be broken (started above zero), because readers perceive bars as proportional to each other and a broken axis distorts those proportions. confirmedas of 2026-11-14

      • Data visualisation: charts · UK Government Analysis Function · Primary source · 2022-05-19 · Bar charts, Do not break the numerical axis on bar charts (retrieved 2026-11-14)
      • Data visualisation: charts · UK Government Analysis Function · Primary source · 2022-05-19 · Bar charts, Do not break the numerical axis on bar charts (retrieved 2026-11-14)
    49. [50]

      The guidance cites an HM Treasury bar chart of inflation whose y-axis started at 8%, which exaggerated a fall from 11.1% in October 2022 to 10.1% in January 2023. confirmedas of 2026-11-14

      • Data visualisation: charts · UK Government Analysis Function · Primary source · 2022-05-19 · Bar charts, description of the Office for Statistics Regulation example (retrieved 2026-11-14)
    50. [51]

      UK government guidance generally advises against dual-axis line charts, saying they are easily misinterpreted and can be manipulated to show almost any relationship between two series. confirmedas of 2026-11-14

      • Data visualisation: charts · UK Government Analysis Function · Primary source · 2022-05-19 · Line charts, Dual axis line charts (retrieved 2026-11-14)
      • Data visualisation: charts · UK Government Analysis Function · Primary source · 2022-05-19 · Line charts, Dual axis line charts (retrieved 2026-11-14)
      • Data visualisation: charts · UK Government Analysis Function · Primary source · 2022-05-19 · Line charts, Manipulation example (retrieved 2026-11-14)
    51. [52]

      UK government guidance on communicating uncertainty recommends showing confidence intervals around estimates where applicable, and warns that changes between periods can fall within survey error and random variation, especially in weekly, monthly or quarterly series. confirmedas of 2026-11-14

    52. [53]

      The UN Fundamental Principles of Official Statistics were endorsed by the UN General Assembly in resolution 68/261 of 29 January 2014. confirmedas of 2026-11-14

    53. [54]

      A federal rule, 5 CFR part 1321, sets out the fundamental responsibilities of recognized US statistical agencies and units; from December 10, 2024 it supersedes any guidance or Statistical Policy Directives to the extent that they conflict with it. confirmedas of 2026-11-14

    54. [55]

      Under 5 CFR part 1321, each recognized US statistical agency must produce relevant and timely statistics, conduct credible, accurate and objective statistical activities, and protect the trust of information providers by ensuring the confidentiality and exclusive statistical use of confidential statistical data. confirmedas of 2026-11-14

    55. [56]

      Each recognized US statistical agency must publicly announce and keep to a release schedule, publishing the dates of its regular releases for the next calendar year before that year begins. confirmedas of 2026-11-14

    56. [57]

      The ONS says seasonal adjustment removes systematic calendar-related variation so that consecutive periods can be compared; for example, retail sales rise each December because of Christmas. confirmedas of 2026-11-14

    57. [58]

      The Census Bureau produces and maintains X-13ARIMA-SEATS, seasonal adjustment software that offers both model-based adjustment using a version of the SEATS software originally developed at the Bank of Spain and nonparametric X-11 adjustments. confirmedas of 2026-11-14

    58. [59]

      BEA says real (chained) GDP is adjusted to remove the effects of inflation so periods can be compared, while current-dollar (nominal) GDP uses market prices of the period measured; the headline GDP growth figure is the change in real GDP. confirmedas of 2026-11-14

    59. [60]

      In Pew's example, a poll with a 3-point margin per candidate shows a 5-point lead, but the margin for the difference is about 6 points, so the true lead could plausibly be anywhere from -1 to +11 points. confirmedas of 2026-11-14

    60. [61]

      Pew says nonresponse and deliberate oversampling of some groups leave raw survey samples as imperfect models of the population, and weighting adjusts them by aligning the sample with known population characteristics. confirmedas of 2026-11-14

    61. [62]

      Public opinion surveys have traditionally been weighted on a core set of demographics (sex, age, race and ethnicity, education and region), which are well measured by large government surveys such as the Census Bureau's American Community Survey. confirmedas of 2026-11-14

    62. [63]

      The Cochrane Handbook also warns against assuming a small p-value means an important benefit; the p-value addresses whether an effect is exactly zero, and in a large study a small p-value may reflect a trivial effect. confirmedas of 2026-11-14

    Revision history (1)
    1. Page created.

    Created Nov 14, 2026. Last reviewed by an editor on Nov 14, 2026. Next scheduled review: Feb 14, 2027.

    Cite this page

    "Statistics and evidence: a crash course." ContentLora, updated Nov 14, 2026. https://contentlora.com/explain/statistics

    Spotted an error? Suggest a correction or emailcorrections@contentlora.com.