“这张图表在骗你！”——数据素养入门
本页标题、字幕、摘要和阅读笔记为已保存的 AI 译文，可能有误。时间戳与来源一致，可展开查看原始字幕，并对照原视频核对。
在这个数据素养入门工作坊中，通过考察数据来源、抽样、语境和视觉选择，学习质疑误导性图表。
文字记录来源：YouTube上创作者上传的字幕。Vidleaf添加了AI标点；保留原始措辞和时间戳锚点。原始文字记录可在下方获取。
AI生成的摘要
这个数据素养工作坊讲解如何在语境中解读数字和图表。它从关于谁收集了数据、为什么收集以及如何呈现的问题开始，然后利用图书馆检索和图表比较将这些疑问付诸实践。示例涵盖抽样偏差、樱桃挑选、相关性与因果关系的混淆，以及误导性的坐标轴或图表类型。目标是形成有依据的怀疑态度，而不是拒绝一切可视化。读者可以利用所链接的段落来考察某一主张背后的证据，并将数据本身与围绕它构建的论证区分开来。
要点
- [7:52] 询问是谁收集了数据、出于什么目的，以及其语境如何影响解读。
- [28:24] 樱桃挑选可以使选定的数据子集看起来支持一个更宽泛的叙事。
- [33:34] 在接受图表的视觉印象之前，检查坐标轴、基线和间隔。
根据文字记录生成。请对照原始视频核查所链接的段落。简要说明和要点由 Vidleaf 于 2026-10-21 审核。
完整概述
教学服务馆员 Russel Peterson 以概述本次数据素养工作坊的目标开场：理解数据是什么，以及数据如何在修辞、商业和政治层面被使用；识别数据呈现中的错误与差异；以适度的怀疑态度接触数据可视化，而不是把每一张可视化都视为天然可疑；以及向他人传达解读。本次课程说明，数据从语境中获得意义——它如何被呈现、如何被收集，以及由谁呈现——并引用 Chantel Risdale 以及 Javier Calzado Prado 和 Miguel Angel Marzal 的定义，这些定义将数据素养界定为收集、管理、评估、批判性评估和合乎伦理地使用数据的能力。参与者练习在图书馆的统计数据库中从 lib.ua.edu 开始进行关键词检索，查找关于美国按制造商划分的 COVID-19 疫苗接种剂次的研究，以及关于高中生尼古丁电子烟使用的研究，并讨论 CDC 数据显示，截至 17 年 2021 月，辉瑞占疫苗接种剂次的明显多数，其次是 Moderna，然后是 J&J，为 15 百万剂，同时还有一项 2020 CDC 研究显示，美国高中生电子烟使用从 2017 到 2018 急剧上升，随后在 2020 略有下降。随后一部分聚焦于数据如何可能以误导性方式呈现，无论是有意还是无意，引用 Darrell Huff 的 1954 年著作《How to Lie with Statistics》，并论证仅凭数字无法讲述故事、提出论点或说服他人，而合乎伦理地使用数据既需要以诚实和理解来使用文字的写作者，也需要能够解读数据并判断其使用是否恰当的读者。误用的例子包括抽样偏差、樱桃挑选、将相关性与因果关系混为一谈，以及误导性可视化，例如截断的 y 轴、省略基线、不一致的间隔、被打破的惯例和被误用的饼图。结尾部分涵盖选择合适的图表类型，以及一项动手的可视化可靠性排序活动，比较一张常规的人口密度地图与一张误导性的英国选举席位 Sun 甜甜圈图，批评一张按年龄和性别划分的上网时长气泡图和一张美国平均家庭债务堆叠条形图（该图相对于抵押贷款夸大了汽车贷款），并将数据素养界定为批判性评估数据如何以及为何被收集、呈现和展示以支持商业、修辞或政治论点的能力，包括检查方法部分并留意樱桃挑选和被操纵的坐标轴；演讲者提出可以回答问题，并说明带有演讲者备注的幻灯片将发送给已注册参与者，录像也将提供，同时邀请未来的观看者通过电子邮件或电话联系。
带时间戳的章节摘要
工作坊开场：数据素养目标
- [0:04] Russel Peterson 以数据素养入门介绍开启本次工作坊。
- [0:14] Peterson 是一名研究教学服务馆员，主要与现代语言系以及顶点项目 Center for Student Success 合作。
- [0:26] 他的研究兴趣包括虚假信息、阴谋论思维以及培养批判性思维技能，尤其是关于数据如何被呈现的问题。
- [1:01] 课程目标：参与者理解数据是什么，以及数据如何被用于修辞、商业和政治目的。
- [1:17] 参与者应能够识别数据呈现方式中的错误与差异。
- [1:23] 参与者应以适度的怀疑态度接触数据可视化，而不是把每一张可视化都视为天然可疑。
- [1:43] 他们应知道何时质疑数据是如何收集的、如何呈现的，以及正在构建什么叙事或论点。
- [1:53] 最后，参与者应能够与同伴及他人交流对数据的解读。
数据如何获得意义与权威
- [2:18] 数据是信息，可以是定量或定性的，通常为特定目的而收集，例如调查。
- [2:58] Pew Research Center 通过电话访谈和焦点小组以及调查回复收集定性数据。
- [3:17] 数据可以是原始的或清理过的，显示在表格或图表中，或表示为数据可视化，例如热力图。
- [3:37] 数据通过其语境获得意义：它如何被呈现、如何被收集，以及谁负责呈现它。
- [4:12] 数据被视觉化呈现，以帮助发现模式和理解，并被用来说服人们接受某个论点。
- [4:34] 在科学中，数据用于检验假设；而在商业中，数据通过传达难以挑战的数字权威来为决策提供依据。
- [5:23] 可以通过对数据呈现方式保持怀疑来质疑数字，例如 20% 台 iPad 销量上升，而整体销量却在下降。
- [5:45] 数据在修辞上传达权威以赢得争论，在商业上用于销售产品，在政治上用于获取权力，例如用民调为某项政策提供依据。
定义数据素养
- [6:37] 数据素养被界定为一种技能，旨在更流畅地处理数据并能够解读数据。
- [6:51] 该领域的图书馆员帮助定义了什么是数据素养。
- [6:57] Chantel Risdale 将数据素养定义为以批判性方式收集、管理、评估和应用数据的能力。
- [7:09] Javier Calzado Prado 和 Miguel Angel Marzal 将数据素养定义为信息素养的一个组成部分，使个人能够访问、解读、批判性评估、管理并合乎伦理地使用数据。
- [7:20] 本次工作坊不会聚焦于收集或使用数据的方法。
- [7:24] 本次工作坊将涵盖如何评估数据、批判性评估数据，以及通过图书馆的一些数据库访问数据。
- [7:44] 上下文中的数据涉及数据是如何收集的以及如何建立上下文。
评估数据上下文的问题
- [7:52] 上下文可以通过提出意义建构问题来确定。
- [7:59] 询问谁是负责收集数据的实体或组织，因为像皮尤研究中心这样受尊敬的研究机构不同于党派性或商业性来源。
- [8:32] 询问收集了什么类型的数据以及正在测量什么，例如定量数据或有多少人回答了一个问题。
- [8:50] 考虑收集方法，例如电话访谈，以及固定电话与手机数据的含义。
- [9:08] 考虑数据是在哪里收集的，以及环境或位置是否影响可信度。
- [9:24] 考虑确切的时间线，因为20年前收集的数据今天可能不再相关。
- [9:32] 询问为什么收集数据以及意图是陈述的还是推断的，包括法律要求的目的，如犯罪数据。
- [10:01] 考虑收集的方法和媒介，因为互联网投票可能存在固有問題。
通过关键词搜索查找统计数据
- [10:42] 学生被要求通过访问统计数据库自行查找数据。
- [10:47] 他们将查找两项特定研究，并从研究问题中提取关键词以应用于搜索。
- [10:59] 示例研究问题：美国按制造商划分的COVID-19疫苗接种剂次。
- [11:13] 可以提取COVID-19疫苗和制造商等关键词并应用于搜索。
- [11:19] 第二个示例：有多少高中生使用尼古丁电子烟弹，以高中生和尼古丁电子烟为关键词。
- [11:34] 演讲者将退出演示，前往图书馆网站并展示如何获取统计数据。
- [11:51] 该活动涉及查找数据集并对其提出意义建构问题。
- [11:59] 意义建构问题包括谁是负责收集数据的实体、数据是什么、何时何地收集、为什么、出于什么目的以及使用了什么媒介。
在图书馆网站上查找CDC疫苗数据
- [12:17] 讲师导航到图书馆网站lib.ua.edu并点击数据库图标。
- [12:24] 她将数据库筛选到“S”部分并向下滚动到统计。
- [12:49] 她确认每个人都到达了同一页面，然后继续。
- [13:09] 她回到幻灯片，并给大家几分钟时间查找研究并分享结论。
- [15:15] 一位参与者找到了一项研究，显示接种的疫苗剂次数量，辉瑞占明显多数，其次是Moderna，然后是J&J，为15百万剂。
- [15:57] 该研究显示了截至17年2021月按制造商划分的疫苗接种剂次。
- [16:40] 数据由疾病控制中心收集，来源链接提供了更深入的收集信息。
- [16:57] CDC从疫苗合作伙伴处收集数据，包括诊所、零售药店和长期护理机构。
- [17:12] 数据包括免责声明，即完全接种疫苗意味着在不同日期接种两剂或接种一剂单剂疫苗。
- [17:43] 由于CDC出于政府目的收集此信息，因此它是引用美国疫苗接种剂次的一个可靠来源。
高中生尼古丁电子烟数据搜索
- [18:25] 小组转而搜索高中生尼古丁电子烟的数据。
- [18:38] 他们询问研究数据是如何以及在哪里收集的，并邀请聊天输入。
- [18:55] 一位参与者指出该研究似乎来自CDC。
- [19:06] 他们进行了一项关于尼古丁电子烟弹、按年级分列的流行率的统计搜索。
- [19:33] 演讲者表示正在查找的研究似乎来自皮尤研究中心，而另一个找到的是最新的2020研究。
- [19:48] 一项2020研究被突出显示，其链接被放入聊天中。
- [20:17] 数据显示，美国高中生使用电子烟的比例从2017年到2018年急剧上升，然后在2020年略有下降。
- [20:33] 来源是2020年底发布的《发病率和死亡率周报》，深入描述了电子烟使用情况。
- [20:54] CDC报告强调了电子烟对健康的影响，尤其是在年轻人中。
评估数据来源和意义
识别误导性数据呈现
数据需要作者和读者
抽样偏差与调查假设
- [23:54] 抽样是一种统计实践，即选择一小群个体或估计值来表征一个总体。
- [24:19] 当样本更能代表整个总体而非一小部分时，可以做出更强的论断。
- [24:31] 当目标总体的成员被过度代表或代表不足，或者数据收集者受到损害时，就会发生抽样偏差。
- [24:54] “10 位牙医中有九位推荐”的口号说明了对调查对象、样本量、受访者资质以及撒谎动机的假设。
- [26:56] 金钱、吸毒或性等敏感话题会导致人们夸大或隐瞒自己的行为。
- [27:17] 一份校友报告声称阿拉巴马大学毕业生平均年收入超过 100,000 美元，这引发了关于调查了多少校友以及哪些校友的疑问。
- [27:49] 一项民意调查称 85% 的美国人从未消费过非法物质，这值得怀疑，因为人们有动机对非法活动撒谎。
作为误导性统计的樱桃采摘
- [28:24] 樱桃采摘是指仅使用一小部分数据来推销某种叙事的做法。
- [28:34] 苹果被引用为一个可能的例子，即一家公司只分享一部分数据以显得好看并推高股价。
- [29:02] 季度产品销售数据可能忽略长达数年的销售下滑趋势。
- [29:13] 关于气候变化和犯罪率的辩论是樱桃采摘的大受害者，因为快速的年度变化不能反映数年或数十年的趋势。
- [29:33] 气候变化否认主义可能展示一小部分气候数据表明变冷，忽略数十年的数据显示温度以不稳定的速率上升。
- [30:14] 《旧金山纪事报》的一条推文称汽车盗窃案上升了 753%，忽略了 2020 的疫情让人们待在室内，而更长期的数据表明犯罪自 1970 年代以来一直在稳步下降。
口罩数据中的相关性与因果关系
误导性可视化实践
- [33:34] 截断图表和省略 y 轴基线是可视化中用于误导的做法。
- [33:41] 将基线设在高数字如 94 百万而不是零，可以使数据显得比实际更夸张或戏剧化。
- [34:11] 一张声称超过 100 百万人接受联邦福利的图表显示从 2009 到 2011 明显激增，但若以零为基线，这将是非常微小的变化。
- [34:32] 该图表统计的是任何居住在至少一人获得项目福利（如医疗补助或贫困家庭临时援助）的家庭中的人。
- [35:20] 一张基线设定在接近五英尺的图表夸大了南非和印度女性的矮小，使拉脱维亚和澳大利亚女性显得巨大，而差异仅几英寸。
- [36:13] x 轴和 y 轴之间的间隔应均匀一致；操纵它们会通过标签或图示元素夸大增加或减少。
- [36:36] 一张计划生育图表显示从 2008 到 2013 癌症筛查和预防服务下降而堕胎上升，但数字（2 百万到 900,000 对比 300,000）似乎不在同一平面上。
- [37:16] 一张关于获得高中文凭的学生的图表显示，从2009年的75%到2015年的82%略有增加，但被操纵的y轴和堆叠的书本使得差异难以判断。
打破数据可视化惯例会误导
- [38:20] 用数据可视化操纵或误导人们通常意味着违背标准惯例。
- [38:33] 一个常见的惯例是，对于折线图和条形图，y 轴应从左下角的零开始。
- [38:44] 图表已变得更加标准化，因此违背惯例乍一看可能会非常令人困惑。
- [39:14] 一张臭名昭著的可视化将 y 轴起点设在 1000 左右，以论证在 2005 年《不退让法》之后枪支谋杀案减少。
- [39:32] 如果 y 轴从零开始，图表将显示在佛罗里达州实施该法律后枪支谋杀案实际上增加了。
- [40:03] 人口密度图通常使用较深的颜色表示人口更密集的地区，例如加利福尼亚、伊利诺伊、纽约、马萨诸塞和新泽西，而怀俄明和蒙大拿则使用较浅的颜色。
- [40:28] 如果反转这一惯例，用浅色阴影表示更密集的人口，会让习惯标准做法的读者感到困惑。
- [40:59] 在社交媒体（如Twitter或Instagram）上快速浏览一个打破惯例的可视化，很容易误导他人。
误用饼图与选择正确的可视化
可视化可靠性排名活动
比较图表可靠性
- [46:38] 展示两个可视化：一个人口密度图和一个英国工党对保守党席位的圆环/饼图。
- [47:22] 人口密度图遵循惯例，使用红色表示较高密度，黄色/绿色表示较低密度。
- [47:45] 《太阳报》的圆环图具有误导性：UKIP没有赢得席位，但看起来至少拥有自由民主党赢得席位的一半。
- [48:07] SNP看起来几乎与自由民主党面积相同，尽管赢得了更多席位。
- [48:29] 另外两个可视化展示了美国普通家庭的债务类型和按年龄和性别划分的上网时间，使用气泡图。
- [49:04] 气泡图是一种区分男性和女性的奇怪方式，使气泡难以辨别。
- [50:11] 气泡图通常展示频率，但在此背景下不合理；条形图会更合适。
- [50:47] 堆叠条形图可能是捕捉债务数据的最佳方式，但有点误导，因为汽车贷款看起来是抵押贷款的四分之三，尽管28,000远小于176,000。
- [51:12] 与抵押贷款债务和学生贷款相比，信用卡和汽车贷款需要减半或显著缩小。
未来的数据素养与批判性评估
Q&A（问答）与幻灯片获取
结束致谢与联系信息
[0:01] 好的。现在你们能看到幻灯片吗？是的，我能看到。好的，太好了。那么欢迎参加本次工作坊——数据素养入门。先自我介绍一下。我叫 Russel Peterson，是一名研究指导服务图书馆员。我主要与现代语言系以及 Capstone 学生成功中心合作。但我的研究兴趣还包括虚假信息、阴谋论思维以及培养批判性思维能力，尤其是当涉及数据如何呈现时——无论是在学术来源中，还是在我们每天接触的信息中，比如社交媒体或其他更传统媒体空间里的信息。这也是我的联系方式，如果有人在本次演示后想联系我。今天这场讲座的一些目标：我希望每一位参与者——无论是现在参与的，还是之后观看的——都能理解
[1:10] 数据是什么，以及它如何被用于修辞、商业和政治目的；识别数据呈现中的错误和差异；并且能够在多种语境中接触数据可视化，同时保持适度健康的怀疑态度——不是把每一个可视化都看作本质上值得怀疑的东西，而是运用一种健康程度的怀疑，知道什么时候适合说：让我们真正质疑一下，比如数据是如何收集的、它是如何被呈现的，以及这里正在构建什么样的叙事或论点。
[1:53] 最后，能够沟通。当你遇到数据可视化或其他类型的数据呈现时，能够与同伴或日常互动的任何人交流对数据的解读。那么，首先让我们谈谈什么是数据，以及它如何被用于日常目的。通过定义它。首先，我可以说数据是信息，可以是定量或定性的形式，而且它通常是为特定目的收集的，所以这可能是从调查中收集的数据，试图衡量，例如，有多少人在使用某个特定商业产品，比如某个化妆品品牌，或者
[2:47] 比如某种技术，比如 iPhone，或任何其他类似技术，所以它可能是为了特定商业目的；它也可能是为了了解人们对某个特定议题的感受而收集的数据，比如 Pew Research Center 就在收集数据，既有定性的，比如人们的电话访谈和焦点小组，也有某种收集人们如果在回答调查问题时在特定情境下可能如何回应的数据。因此，这些数据可能是原始的，也可能是清洗过的；它可能以某种方式组织，并以表格形式展示，比如通过 Excel 电子表格中的图表，或者它也可以被表示为数据可视化，你可以在这里右侧看到。
[3:37] 这就像一张东欧的热力图。因此，数据是通过它所处的语境、它如何被呈现或如何被获取或收集而获得意义的，所以无论是谁在收集这些数据，它是通过图表来呈现，还是通过地图来呈现，等等，真正赋予数据意义的不只是它提出了什么样的问题，还有它如何被置于语境中，比如谁负责呈现这些数据；数据被视觉化呈现，以帮助发现模式、促进理解，它当然也被用来说服人们，比如让人们同意某个论点，或提供清晰的目的。然后使用数据，所以在科学界
[4:41] 数据被用来检验假设。但在其他领域，比如商业，就像我之前说技术时提到的，数据被用来为决策提供正当性，通过传达一种难以挑战的数值权威。比如 Apple 发布一组报告说，哦，我们所有产品的销售额在上个季度增长了 20%。
[5:11] 表面上看这很难挑战，因为数字本身可能就是一种难以争辩的权威，除非你知道更多。除非你更加——我会说——对数据如何被呈现持怀疑态度，比如如果他们说，哦，我们 iPad 的销售额在上个季度增长了 20%。但随后你看到在一段时间内整体销售额在下降，这就会让你质疑他们实际上是如何呈现这些数据的、如何传达这些数据的。因此，数据被用来传达权威的三种主要方式之一是修辞性地使用。它被用来赢得某种争论。这在政治领域确实如此，或在其他商业领域，用来销售特定产品。但在政治领域。它也被用来获取权力。所以如果人们正在发布
[6:16] 某些民调来表明，比如，哦，75% 的美国人在这个特定议题上意见一致，或者他们不喜欢这项特定政策，这被用来为首先制定那项政策提供某种正当性。那么什么是数据素养，我们如何理解这项特定技能——它被用来让人对数据更熟练、能够解读数据。
[6:51] 我在这里对帮助定义了在我们特定领域中数据素养是什么的图书馆员说过，Chantel Risdale说数据素养是收集、管理、评估和应用数据的能力，并且是以批判性的方式。而Javier Calzado Prado和Miguel Angel Marzal将数据素养定义为信息素养的组成部分，使个人能够访问、解释、批判性评估、管理、合乎伦理地处理和使用数据。对于今天的研讨会，我们不会真正讨论收集或使用数据的方式，但我们肯定会讨论如何评估数据、如何批判性地评估它，以及我们如何通过图书馆这里的一些数据库来访问它。
[7:44] 那么，上下文中的数据，数据是如何收集的，因此我们如何建立上下文并赋予数据意义，以及我们确定上下文的方式是提出一些意义建构问题。例如，谁是负责收集数据的实体组织，这可能很重要，如果它是像皮尤研究中心这样备受尊敬的研究机构，还是更像党派来源，或者更偏向商业来源，这可能会让你质疑，比如这些数据是如何收集的，我们应该信任背后的实体吗。
[8:32] 收集了什么类型的数据，测量的是什么。所以这是定量数据吗，这只是数字吗。这就像有多少人回答了某个特定问题，或者这是，比如，这些数据是通过电话访谈收集的吗。那么，比如说，通过现在很少有人拥有的固定电话收集的数据与通过手机访谈收集的数据相比，有什么影响，例如，数据是在特定环境或地点收集的，如果只是在美国收集的数据，这相关吗，如果数据来自两个非常不同的地方，这如何影响数据是否可信。
[9:24] 什么时候，比如收集这些数据的确切时间线是什么，数据可能是20年前收集的。这是否意味着它今天仍然相关。为什么，收集这些数据的目的是什么。这些意图是声明的还是被推断的。所以任何正在收集的数据，或任何特定目的，无论是法律要求的，也许比如犯罪数据必须依法收集，他们是如何收集的。
[9:56] 所以你也必须考虑数据收集背后的目的。然后是的，收集数据的方法是什么，以及采用什么媒介。我认为互联网投票可能有很多固有的问题。因此，重要的是要看看数据是以什么方式收集的，这样你才能正确评估，好的，这个来源是否恰当地评估了特定媒介在以特定方式收集数据时呈现的所有独特的不同挑战。现在我要让你们。自己查找数据，我们将通过访问统计数据库来做到这一点，我们将查找两项特定的研究，通过这样做，我们只是从这些研究问题中提取关键词并应用这些关键词
[10:59] 在搜索和城市东西中，那么按制造商划分的美国COVID-19疫苗剂量接种数量是多少。我们可以通过提取关键词如COVID-19疫苗和制造商并将这些关键词应用到城市中来实现，同样对于有多少高中生使用尼古丁电子烟弹。高中生可能是与尼古丁电子烟混合的关键词。所以是的，我将退出这个演示，以便我们可以去图书馆的网站，并实际向你们展示如何访问统计数据。
[11:42] 我看到有新的人加入，你好，我们现在要开始查找某些数据集的活动，我们将提出一些意义建构问题。谁是负责收集数据的实体。数据是什么。何时收集的。在哪里收集的，为什么，出于什么目的，以及用什么媒介收集的。所以我要向大家展示如何实际访问城市东西，所以我在图书馆的网站上，即lib.ua.edu。现在在主页上，我将点击数据库图标。然后我将筛选到“S”数据库。然后向下滚动到统计。
[12:38] 好的，我们开始吧。现在，测试。好的，大家都能到达我现在的位置吗。是的。是的。好的，太好了。看到一个竖起大拇指。太好了。好的，所以我将回到这里的幻灯片。然后就像给大家几分钟时间查找这些研究，然后与我们分享你们的结论。
[13:57] 感谢聊天，放下一些关于的信息。所以我认为你确实找到了那种研究
[15:15] 就像概述的那样。辉瑞公司接种的疫苗剂量数量绝对是明显的多数，其次是Moderna，然后是J&J 15百万。是的，让我去城市并展示你们那个研究。看到那个场景剂量制造商。截至2018月按制造商划分。好的。
[15:50] 好的，所以这就是这项确切的研究。截至17 2021，这些是按制造商划分的已接种疫苗剂次数量。就理解性问题而言，关于这项研究以及数据是如何收集的，你们能够得出什么结论？所以是由CDC散布的，对吧，是的，我们可以从来源看出。它是由美国疾病控制与预防中心收集的。
[16:45] 如果你点击他们的来源链接，他们甚至会给你更深入的信息。有点像告诉你他们是如何能够收集这些信息的，所以他们能够从所有疫苗合作伙伴那里获取这些数据。这包括诊所、零售药店、长期护理机构。让我们看看那里甚至还有一些免责声明，说明完全接种疫苗是指在不同日期接种两剂，或者接种一剂单剂疫苗，所以这些数据来源包括发布者。你可以自定义，所以因为这是CDC，他们有点像在为政府目的收集这些信息，那
[17:50] 就像是他们的职责，那是他们的使命，有点像追踪每个美国人的健康结果，尤其是在疫情期间。追踪这些信息是他们的工作。所以你可以基于这一点来访问，这说明在引用美国正在接种多少疫苗时，这是一个值得使用的来源。好的。
[18:25] 然后对于。让我们看看高中生尼古丁电子烟，在我进行搜索之前。是的，我想人们在这项研究中发现了什么。就数据是如何收集的以及在哪里收集的而言。现在，请随意把它放在聊天中或直接发言。好的。好的，这一项看起来也来自CDC。是的。
[19:06] 好的。那我们再做一次统计搜索，尼古丁，尼古丁电子烟弹。好的，比如说学生中尼古丁烟弹电子烟的流行率以及美国按年级划分的情况。好的。所以你可能会找到另一项CDC研究。但我在这里查找的这项似乎来自皮尤研究中心。但还有一项，尽管它是2020的最新研究。好的，所以你找到了一项2020研究，太好了，可以放上来，我会把链接放上去。好的，所以我们也可以，是的，看看那个设置，尼古丁电子烟2020。我把链接放在聊天里了。好的，太好了。
[20:09] 与我们分享高中生的情况。好的，太好了。新数据，美国高中生使用电子烟的百分比，我们可以大致看到从2017到2018的急剧上升，然后在2020略有下降，可以具体点击来源链接，看到这是更，更生动，抱歉，是发病率和死亡率周报，于2020末发布，有点像对电子烟使用情况给出了更深入的描述。
[20:50] 再说一次，这就像是CDC的职责。有点像强调某个特定产品比如电子烟的健康影响，尤其是在年轻人中。好的。太好了。所以回到我们的演示，我认为培养这些技能是值得的，即尝试弄清楚数据来自哪里。
[21:29] 这样你就能正确评估，好的，这些数据的含义是什么？它来自哪里？因为它在那些非常方便的图表中往往可能没有被呈现出来，你看到的统计数据可能是你在网上、Twitter上看到的一张图表截图，或者它可能以多种不同方式呈现。所以即使数据来自某个地方。你也必须注意它是如何被呈现的，以及它是否被用来以某种方式误导人。
[22:08] 这就进入了我们演示的下一部分，即能够识别出无论是故意还是无意，有些数据以某种方式呈现出来是具有误导性的，而这句话有点像这个主题上的一篇开创性文章。我要引用达雷尔·哈夫的话，他在1954写了《如何用统计说谎》。统计方法和统计术语对于报告社会和经济趋势、商业状况、民意调查、人口普查的大量数据是必要的，但如果没有诚实、理解地使用这些词语的作者，以及知道它们含义的读者。那就只能是语义上的胡说八道。所以，数字本身并不能真正告诉你一个故事，不能真正告诉你某个论点或以某种方式说服你，只有作者才能合乎伦理地使用这些数据并传达。
[23:11] 也许是某种叙事，或者。它也需要读者能够理解、能够解读这些数据，以便理解它，也许还能运用他们的判断力，说这是对数据的恰当使用，或者这是不恰当的使用，或者某人在传达信息时没有坦诚。所以我将大致介绍几个不同的例子，说明数据如何以这种方式被滥用，以及如何被用来误导人。
[23:50] 所以第一个例子是一种叫做抽样偏差的东西。这是一种统计实践，有人试图本质上这是一种选择一小群个体或估计某个总体某些特征的方式，所以会提出一些主张。当样本更能代表整体时，会提出更强的主张，而不是只取这些个体的一小部分，所以当预期总体的成员被过度代表或代表不足时，就会发生抽样偏差，当收集数据的人以某种方式受到损害时也会发生。所以，例如，我认为一个常见的例子
[24:54] 抽样偏差就像你在很多牙科产品上看到的标语，比如“10 位牙医中有九位推荐使用高露洁这款牙膏”。这里面隐含着某些假设，比如：哦，那一定意味着成千上万、甚至数万名牙医都接受了这项调查。然后所有这些牙医肯定都回复了那份调查，而且如果他们自愿这么做的话，肯定没有动机对所有人撒谎。
[25:36] 但是，当你遇到这样的陈述时，其中隐含着好几个假设：调查对象是谁，样本量是多少，所以有可能只调查了 50 位牙医。那么这绝对不足以代表整个牙医行业，而且这些特定的受访者是谁，他们真的是牙医吗，他们是否被迫出示自己的资质证明，还是说只是随便在街上找到的人，然后说，哦，你是牙医专业人士。还有，受访者是否有撒谎的动机。例如，他们是否因为给出意见而获得报酬，还是说这是某种特定的。
[26:31] 他们。我该怎么表述呢。回复一项调查对他们的职业没有任何好处，也许如果调查来自非常有声望的机构的话。所以这些就是一些例子，说明为什么你可能需要考虑：人们是否有动机不坦诚。我认为这也适用于当人们被问及非常敏感的话题时，比如关于金钱、毒品使用或性的话题。这些话题往往是人们倾向于夸大或隐瞒自己行为或活动的话题。
[27:17] 所以抽样偏差的一些例子可能像是这样一份旧的校友报告，说阿拉巴马大学毕业生的平均年收入超过 100,000 美元。所以值得问一问：调查了多少校友，是只调查了商学院学生吗，是只调查了工程学院学生吗，总共调查了多少阿拉巴马大学毕业生。然后另一个例子是，一项新民意调查称 85% 的美国人从未消费过非法物质。
[27:55] 人们有动机在这些话题上撒谎，因为他们可能不想在调查中承认做了任何违法的事情，他们可能不信任调查者。所以值得考虑的是，从某项调查中收集的样本可能会在某个特定方向上存在偏差。人们用来可能误导他人的另一种方法，另一种统计方法，是樱桃采摘，樱桃采摘就是只使用一小部分数据的做法。来推销某种特定叙事。所以我想我之前举了苹果作为例子。也许是一家公司只。他们可能只想分享特定的一部分数据，让自己看起来不错，也许还试图推高自己股票的市场份额，说，就像这样。
[29:02] 这个特定季度，这款科技产品卖出了这么多台，但它没有看这款产品多年来销售不佳、销量实际上在下降的长期趋势。因此，关于气候变化和犯罪率等问题的争论是樱桃采摘的一大受害者，因为逐年的快速变化并不能反映多年或数十年尺度的趋势。
[29:33] 所以，很多参与气候变化否认主义的人，会展示一小部分气候数据，说，哦，实际上这表明地球在随着时间推移而变冷，或者只是有太多快速变化，你无法真正说地球在变暖。但这没有考虑到数十年尺度的数据，这些数据显示地球温度实际上正在以不稳定的速度上升，并且正在超过某些工业化水平。犯罪数据也是如此。
[30:14] 这是《旧金山纪事报》的截图。几个月前他们发了一条推文，说汽车破窗盗窃案上升了 753%，这就引出了一个问题：2020 年到底发生了什么。当时有一场大规模疫情，导致很多人待在家里。所以这也许只考虑了五年或 10 年的数据，而这些数据可能显示犯罪。
[30:47] 实际上可能在下降，有很多数据表明犯罪率多年来仍在稳步下降，尤其是自 1970 以来的几十年里。另一种人们用来误导的统计方法，我想，是把相关性和因果关系混为一谈。仅仅因为，我相信你们俩以前都听过这个说法，仅仅因为某件事是相关的，或者两件事是相关的，并不意味着其中一个是另一个的原因。所以，即使某个说法最终被证明是真的，就像这个例子中那样。那么，为了描述一下这里的图表，它显示的是没有立即实施口罩强制令的国家病例出现了巨大激增。而实施了口罩强制令的国家病例较少。
[31:50] 这是一个两件事相关的例子。可能有各种各样不同的因素可以解释这种相关现象。例如，其中很多国家，韩国、日本、新加坡，这些都是东亚国家。也许这并不能说明什么。也许是那些国家的某些特定做法，与口罩无关。也可能是病毒在这些社区中并不那么流行，而在这些欧洲和北美国家中则更为流行。
[32:30] 还有很多其他不同的因素可以解释这一点，即使最终发现口罩对遏制冠状病毒的传播非常有帮助。重要的是要意识到这一点，并对那些声称这单一问题就是这个非常复杂问题的原因的论点保持怀疑。因此，值得质疑某些事情，即使最终发现那种情况在某些情况下可能是真实的，口罩非常有帮助，并且碳描述但并不能解释为什么冠状病毒病例在某些地方上升而在其他地方不上升的多方面复杂性。
[33:22] 然后还有一些可视化被用来误导人们的方式，以及在可视化中使用某些做法来误导人们。所以一个例子是截断坐标轴的图表和省略y轴上的基线。所以，在这两个例子中，基线都处于一个非常高的数字，比如在这个例子中，94百万是截止点，而实际上应该是零。这样做的效果是，它可以将数据歪曲得比实际更夸张或更戏剧化。而且，是的，所以右边的图表说现在有超过100百万人在接受联邦福利。
[34:16] 这有点像看从2009到2011期间病例随时间的爆炸式增长。但如果我们将基线设为零，你可以看到与实际数字相比，那只是一个非常微小的变化。我认为那个图表将任何接受联邦福利的人计为任何居住在家庭中的人，并且至少有一人获得了项目福利，可能是Medicare、Medicaid，或者抱歉，任何东西。这里统计的是经过经济状况调查的福利，不是社会保障或Medicare。好的，所以可能是某个家庭中的任何人受到影响，所以可能是Medicaid或其他联邦援助项目，比如。
[35:06] 例如，贫困家庭临时援助。所以任何接受该福利的人。他们将家庭中的每个人都计为接受该福利，所以对于那个特定的图表，这只是需要记住的事情。而左边的那个。这个图表有很多问题，但一个是它有点夸大了南非、印度女性的矮小，因为基线设置在非常接近五英尺的地方，所以这不是等间距的。
[35:47] 所以它有点让人觉得来自拉脱维亚和澳大利亚的女性与来自南非和印度的女性相比是巨大的，而实际上只是几英寸的差别，并不是巨大的边际或显著差异。另一个例子只是对白轴的更多操纵。所以x轴和y轴之间的间隔应该均匀一致，而操纵这些间隔的可视化通过使用标签或彩色插图元素来夸大增加或减少。所以，就像左边这个例子，如果你们能看到的话，是癌症筛查和预防服务在Planned Parenthood下降，但堕胎从2008上升到2013。
[36:52] 但问题是这两个数字应该完全不同。这就像从2百万到900,000，但对于某些检查。出于某种原因，300,000次堕胎或者，或者像这个数字以上，所以数字甚至似乎不在同一个平面上，这就像对白轴的操纵。而对于右边这个例子，在奥巴马总统领导下，比以往任何时候都有更多的学生获得高中文凭。尽管这只是从75%年的2009到82%年的2015的边际增长。
[37:38] 它有点显示这就像超过这个特定数字的一半，基于堆叠的书籍数量。所以，就像，因为基线。因为y轴在这里被操纵了。你无法真正看到获得高中文凭的学生人数的差异实际上有多么深刻。
[38:09] 我会说最臭名昭著的违背例子之一。有点像是用数据可视化操纵人们或误导人们，是违背惯例，违背数据可视化的标准惯例，比如y轴应该从左下角开始，应该从零开始。这就像是一个相当常见的惯例。当涉及到折线图和条形图时。
[38:44] 原因是图表和图形已经变得更加标准化，而违背这种标准化、违背那些惯例，可能会让人乍一看非常困惑，所以普遍认为你应该把y轴从零开始放在左下角，这样就不会让人困惑。而这种臭名昭著的可视化将y轴从1000开始，以传达一个论点，即在2005之后。
[39:23] 在《坚守阵地法》之后，使用枪支实施的谋杀案数量实际上减少了。但问题是，如果你实际上把这个翻转过来，应该为零的地方实际上在这里。你可以看到，在佛罗里达州实施该法律后，使用枪支实施的谋杀案数量实际上增加了。所以，这就像一个例子，说明违背惯例如何误导人们，因为它没有正确。
[39:55] 从正确的起点开始设置y轴。第一，这里的这个例子实际上遵循了惯例，即人口密度图。通常较深的颜色意味着某地人口更密集，所以加利福尼亚州、伊利诺伊州、纽约州、马萨诸塞州、新泽西州的人口密度高于怀俄明州和蒙大拿州。但如果这一惯例在某种程度上被颠倒了。如果人们，例如你试图说浅色阴影表示人口更密集的地方。
[40:41] 那会让已经习惯于知道浅色阴影的地方应该人口密度更低的读者感到困惑，所以他们可能会把某些地方误认为比实际更受欢迎。所以当你看到社交媒体上，比如Twitter或Instagram上的可视化图表，你只是匆匆瞥一眼，并没有真正深入研究时，这一点可能相当重要。如果它违背了某种惯例，就很容易误导别人认为某个观点是真的或不是。然后，我用错了图表，这可能是误导人们或至少让他们困惑的一个相当大的方式，例如饼图就是出了名地被滥用。它们本应表示整体的各个部分。
[41:34] 所以左边这个例子对于尝试过大麻的美国人来说并不太合理。51%说今天，43%说去年，34%说1997。是的，这根本说不通，因为如果你想衡量大麻使用情况随时间可能发生了怎样的变化，你会使用一种非常不同的可视化方式来展示它，而不是饼图，饼图应该总是加起来等于100%，而且它应该是整体的各个部分，它应该展示变量如何表示，或者它与变量的关系，以及它如何表示整体的各个部分，而不是某些其他类型的可视化。有与某些类型可视化相对应的数据，所以有些数据是用来比较两个值的，比较差异
[42:40] 或相似之处，而那些值会使用比较，所以像条形图、哑铃图对于试图衡量关系的人来说相当不错，热力图。例如，你试图展示数据中的联系，或描绘相关性或缺乏相关性，热力图或散点图很适合这一点。分布方面，展示频率或可能随时间变化的可视化，直方图对于这类数据会好得多，这种展示随时间变化的图，相比饼图。
[43:20] 好了，现在我们要对可视化图表进行排名。那么，既然我们得让你们在这里，我会在聊天中放两个PDF。你们可以随意选择一个PDF而不是另一个，但基本上你们要对可视化图表进行排名，从哪个可视化最可靠到哪些最不可靠，就运用你们今天学到的关于可视化如何呈现的知识。
[44:00] 人们可能如何滥用某些可视化来传达论点。你可以看到哪些可视化会更可靠，哪些可能最不可靠。所以我要停下来了。现在我要做的是把这个放到聊天里，让每个人都能看到。看看我能不能在这里分享一个PDF。
[44:48] 稍等一下。我想我要做的是，我觉得我实际上无法在聊天中分享文件。抱歉。所以我要做的是对其中一个包含所有这些可视化图表的文件进行屏幕共享，让你们大家都能看到那个文件，并能够评论哪个最可靠，哪个最不可靠。
[45:19] 让我去找一下那个文件，然后我会分享给大家。我要用第二个。好的，我再分享一次我的屏幕。好的。
[46:08] 你们大家现在能看到这个PDF。如果能看到就给我点个赞。太棒了。是的，很好。那么，呃，让我们看看这里的这四个可视化图表，我要把它缩小。好的，让我们看看。
[46:38] 好的，我们这里有两个可视化图表，一个是人口密度图，另一个是环形图或另一种类似饼图的图表。它展示了英国工党有多少席位，对比保守党有多少席位，根据你们对这些可视化图表的第一印象。你们会如何描述它们的可靠性。一个似乎肯定比另一个更可靠。
[47:22] 它肯定遵循了人口密度图的惯例，红色被用来表示人口密度更高的地方，而黄色和绿色等颜色则表示较低，然后是的，《太阳报》的那个。我觉得要注意的一点是，UKIP党虽然没有赢得任何席位，但在这个图表中显示它至少拥有自由民主党所赢得席位的一半，而SNP几乎拥有与自由民主党完全相同的面积，尽管他们赢得更多。
[48:11] 所以这肯定是误导性的，这里的呈现方式，即使这个政党根本没有赢得任何东西，它却不知怎么地在这个环形图上被表示出来了。是的，UKIP甚至不应该被显示出来，它在这里是零，没错。好的，然后这里的这两个其他可视化图表。一个是展示美国家庭平均的债务类型。然后是底部这里的第二个图表。
[48:45] 它展示了我们按年龄和性别划分的在线消费时间。让我们使用气泡图。一个比听到更可靠。是的，我会说，是的，第二个，这就像一种奇怪的方式，通过展示来区分男女之间的差异。是的，即使是在比较某物与另一物时，也很难辨别这些不同的气泡代表什么。
[49:25] 是的，所以也许12岁的男性比12岁的女性上网时间更多或更少，一开始可能很难衡量。是的，我认为圆圈的不同大小是为了传达那些。这是为了表明更多人在网上花费时间，是的，甚至很难衡量这到底有多相关。
[50:04] 是的，气泡图相当不清晰。是的，气泡图通常用于展示频率。例如，我认为热图可以显示。如果一个地方人口更密集，或者更多人响应某个特定问题，气泡会扩大，变得更大。但在这种背景下，这并不太合理。
[50:37] 你会认为也许条形图在那里更合适，而这张图。是的，我会说这可能是用堆叠条形图来捕捉数据的最佳方式。不过它有点误导，汽车贷款被视为抵押贷款的四分之三，尽管28,000肯定远小于176,000。
[51:12] 所以我认为这里的图表需要在信用卡和汽车贷款方面进行调整，与抵押贷款债务尤其是学生贷款相比，它们需要减半，显著缩小。好的，所以我们将回到幻灯片，问一些关于你未来如何与数据互动的最终想法。现在你已经看到了这些数据如何被用来误导的例子，了解了背景的重要性
[52:18] 数据是如何收集的，如何被呈现和展示以支持特定论点，无论是商业修辞还是政治目的。是的，我只是认为，能够具备数据素养意味着能够批判性地评估它，并能够知道例如。为什么数据以特定方式呈现以及出于什么目的。所以我希望在这次演示之后，你能得到这样的收获。你可以正确地。
[52:55] 看到数据，无论是通过社交媒体还是其他渠道，例如学术来源，并能够也许查看方法论部分，看看他们是如何真正收集这些数据的，我们能否正确信任。传达给我的论点。所以，是的，我希望大家在观看今天的演示后能有这样的收获。但我很乐意回答你的任何问题。让我在这里分享我的引用页面，并分享我今天是如何整理这些内容的，但我很乐意回答任何问题。好的。
[53:46] 在聊天中...我会小心注意意图传达的信息，也注意诸如挑选数据、操纵坐标轴之类的事情。是的，绝对。幻灯片会提供吗？是的！幻灯片包括我在那些幻灯片上的笔记，我将通过电子邮件发送给每位注册参与者，本次演示的录音也将提供。好的！太好了！感谢大家参加！
[54:29] 如果你以后观看这段录音，请随时给我发电子邮件。我会公布我的联系信息，以便大家都能看到。你可以通过电子邮件rtpeterson1@ua.edu联系我，或者通过电话联系我。我很乐意以任何你想要的方式回答你的问题。所以，谢谢大家！我真的很感激你们主动联系并聆听我今天的演示。谢谢！
查看英文原始字幕
[0:01] All right. And can you see the slides right now. Yes, I can. Alright. Excellent. So welcome to this workshop an introduction to data literacy. And just to introduce myself. My name is Russel Peterson, I am a research instructional Services Librarian. I work mostly with the modern languages department and the capstone Center for Student Success. But I also have research interests in disinformation conspiratorial thinking and cultivating critical thinking skills, especially when it comes to how data is presented, both in scholarly sources but also just like the information we interact with every day like on social media or within other more traditional media spaces. And that is also my contact information if anyone wants to get ahold of me after this presentation, some goals for today's session, I want every participant who is either. Participating now, or watching later to understand
[1:10] what data is and how it's used for rhetorical commercial and political purposes, identifying the errors and discrepancies and how data is represented as well as being able to engage with data visualizations in multiple contexts with a healthy degree of skepticism not seeing every single visualization as something to be inherently skeptical about but to exercise, a healthy degree to know when it's the appropriate time to say let's actually question like how the data was gathered how it's being represented. And what kind of narrative or arguments are being constructed here.
[1:53] And then lastly, being able to communicate. Interpret interpretations of data with your peers or with anyone else that you interact with on a daily basis when you come across, either data visualization or some other type of representation of data. And so let's first talk about what is data and how it's used for everyday purposes. By defining it. First, I can say the data is information, either in a quantitative or qualitative fashion and it's often collected for a specific purpose so this could be data that is collected from a survey, and it's trying to gauge, for instance, how many people are using a particular commercial product, say like a makeup brand or.
[2:47] Let's see like a certain technology like the iPhone, or any other type of technology like that so it could be for specific commercial purpose it could be data that's gathered to kind of glean how people feel about a particular issue so the Pew Research Center for instance, is gathering data that is both qualitative like people's phone interviews and focus groups but it's also kind of a gathering, how people might respond in a particular situation if they're answering a survey question. And so that data could be raw or could be cleaned as well it could be organized in a certain fashion and displayed in a tabular way like through charts in Excel spreadsheet or it could be represented as like a data visualization and you can see on the right hand side here.
[3:37] It's like a heat map of Eastern Europe. And so data is given meaning by the context of it's situated in how it is presented or how it is gleaned or gathered so whoever is gathering that data is it being represented through a chart or is it being represented through a map, or what have you, that really is what it's giving the data, meaning not just what kind of questions it's asking but how it's kind of situated as well like who is responsible for presenting this data and data is rendered visually to aid in pattern finding comprehension, it's definitely used to persuade people as to kind of like to have people agree on a certain argument or to give like a clarity of purpose. And then using data so in the scientific community
[4:41] data is used to probe a hypothesis. But in other realms, business, like I was saying earlier with technology data is used to justify decision making by conveying a numerical authority. That's difficult to challenge so it's like Apple for instance is releasing a set of reports saying that that like, oh, like sales of all of our products have been going up by like 20% in this past quarter.
[5:11] It's hard to kind of like challenge that on its face because numbers by themselves can be an authority that can be hard to dispute unless you know something more. Unless you're more. I would say skeptical about how that data is being represented like if they're saying oh like our sales for the iPad have been going up 20% this past quarter. But then you see like over a set of time that overall sales are declining and that makes you question, how they're actually presenting that data and how they're conveying that. And so the three main ways that data is used to convey authority is rhetorically. It's used to win an argument of some sort. This is definitely true in the political sphere, or in other spheres commercial to sell a particular product. But in the political sphere. It's also used to gain power. So if people are like having are releasing
[6:16] like certain polls to indicate that like oh 75% of Americans are in agreement on this particular issue or they do not like this particular policy that is used to kind of justify putting in that pol icy in the first place. And so what is data literacy like how do we like make sense of this particular skill that is used to kind of like have a more fluency with data being able to interpret it.
[6:51] And I've said it to librarians here who helped define what data literacy is in our particular field, and Chantel Risdale says data literacy is the ability to collect manage evaluate and apply data and a critical manner. While Javier Calzado Prado and Miguel Angel Marzal defined data literacy as the component of information literacy, that enables individuals this access interpret critically assess manage handle ethically use data. For today's workshop we won't be really talking about ways of like collecting or using data but we'll definitely be talking about how to evaluate data, how to critically assess it, as well as like how can we access it through so me of the databases here at the Library.
[7:44] And so, data in context, how was data gathered so how can we establish context and give data, meaning for what it is and ways that we can determine context is by asking some sense making questions. For example, who what is the entity organization responsible for gathering the data that could be important, if it is like say a well respected research body like the Pew Research Center, or is it like a more partisan source or is it a more commercially focused source that could lead you to question, like how is this data gathered should we trust the entity behind it.
[8:32] What's what kind of data has been collected what is being measured. So is this quantitative data is this just by the numbers. This is like how many people have responded to a particular question maybe, or is it, like, what's this data gather to like telephone interviews. And what are the implications of, say, data that's gathered through a landline that not many people own anymore versus data that's gathered through cell phone interviews for example, where where's the data being gathered in a particular setting or location, is it relevant if it's just data gathering United States if its data gathered from two very different places, how does that play into whether the data is trustworthy or not.
[9:24] When when is like the exact timeline for collecting this data, the data was collected maybe 20 years ago. Does that mean that it's still relevant for today. Why, for what purpose was this data gathered. Are these intentions stated are being inferred. So any data that is being gathered, or any particular purpose, whether that is like required by law, maybe if, like crime data for instance has to be gathered by law, how are they gathering that.
[9:56] So you have to kind of consider like the purpose behind that data gathering as well. And then yeah, what was the method for collecting the data and what medium to take place. I think internet polls can have a lot of inherent problems. So it's important to kind of see what manner was was data gathered so that you can properly evaluate okay is this like a source that that properly kind of like assessed all like the unique different challenges that a particular medium presents when you are gathering the in a particular way. And now I'm going to have you. Kind of like, look for data on your own, and we're going to do this by going to the statistics database, and we're going to look up two particular studies, and by doing this, we're just going to extract keywords from like these research questions and apply those keywords
[10:59] in a search and city stuff so what are the number of COVID-19 vaccine doses administering us by manufacturer. We can do that by just extracting keywords like COVID-19 vaccine and manufacture and applying those keywords into cities, and the same for how many high school students are using nicotine vaping cartridges. High school students could be a key word mixed in with nicotine vaping as well. So yeah, I'm going to exit out of this presentation here so that we can go to the library's website and can show you actually how to get to statistics.
[11:42] And I see we have a new person enter Hello, we are just now going to starting this activity of looking up certain data sets, and we're going to ask some sense making questions of. Who is the entity that is responsible for gathering the data. What is the data. When was the gathered. Where was it together why, for what purpose, and what medium was it gathered. So I'm going to show everyone just how to actually get to the city stuff so I'm at the library's website which is lib.ua.edu Now on the homepage here I'm going to click the databases icon. And then I'm going to filter down to the "S" databases. And then scroll down to statistics.
[12:38] All right, here we go. Now, to test. All right, was everyone able to get to where I am right now. Yeah. Yes. Okay, great. See a thumbs up. Excellent. Alright so I'm just going to go back to the slide right here. And all just like allow a few minutes for everyone to kind of like look up these studies and then share your conclusions with us.
[13:57] Thank chat, put down some information about. So I think you did find the study that kind
[15:15] of like outlined. What are the number of vaccine doses administered Pfizer's definitely the clear majority, followed by Madonna and then J&J 15 million. Yeah, so let me go to cities and show you that study. See that scene doses manufacturer. By October 2018 by manufacturer. Okay.
[15:50] All right, so this is the this is the exact study. As of October 17 2021, these are the number of vaccine doses administered by manufacturer. And in terms of like sense making questions. What were you able to like determine about this study and how the data was gathered. So scattered by the CDC right so yeah we can see like from the source. It was gathered by Centers for Disease Control.
[16:45] If you click on their Source Link they even give you like more in depth. Kind of like telling you how they were able to collect this information so they were able to get this data from all their vaccine partners. So that includes clinics retail pharmacies long term care facilities. And let's see happen there even, like, some giving some disclaimers accounts we was fully vaccinated receive two doses on two different days, or a seed one dose of a single dose vaccine happen so this data sources so including posters. And you can customize so it's like since it's the CDC like they're kind of collecting this information for governmental purposes, that
[17:50] is like their charge that is their mission to kind of like track the health outcomes for every American, especially during the pandemic. It is their job to keep track of this information. So you can kind of like access like based on that that this is kind of like a worthy source to use when citing about how many vaccines are being administered in the United States. Okay.
[18:25] And then for the. Let's see the nicotine vaping high schoolers up before I do the search. Yeah, I guess like what were people finding about this study. In terms of like how was the data gathered and where it was gathered. Now, feel free to put that in the chat or to speak up. Okay. All right, so this one also looks like it was from the CDC. Yeah.
[19:06] Okay. So do another statistic search nicotine Nicotine vaping cartridges. Okay, so let's say prevalence and nicotine cartridge vaping one students and us by grade. Okay. So you might have found another CDC study. But the one I'm looking up here seems to be from the Pew Research Center. But there was one more although it's the latest study of 2020. Okay, so you found a 2020 study, excellent, to have on, I'll put the link. Okay, so we can also yeah look at that setting nicotine vaping 2020. I put the link in the chat. Okay, great.
[20:09] To share with us high school students. Okay, great. New episodes percentage of us high school students using e cigarettes and we can kind of like see a sharp jump from 2017 to 2018 and then a slight dip in 2020 can go to like the Source Link specifically, and see this is the more ability, more vivid, sorry, morbidity and mortality weekly report that was released at the end of 2020, kind of like giving a more like in depth description about the use of e-cigarettes.
[20:50] And again, that is like the charge of the CDC. Kind of like highlight like what are like the health impacts for a particular product like e-cigarettes, especially among younger people. All right. Excellent. So going back to our presentation here I think it's worthwhile to kind of develop these skills of trying to figure out like where data is coming from.
[21:29] Just that you can properly assess like okay like how, what is the meaning of this data. Where's it coming from because it can be often not represented in those like very handy charts, have you seen statistics might be like a screenshot of a graph that you see online, on Twitter, or it could be represented in multiple different ways. And so even if data is coming from a certain place. You also have to pay attention to how is the state of being delivered and is it being used to mislead in a certain way.
[22:08] And that goes into the next section of our presentation is being able to spot whether intentionally or unintentionally there is data that's being presented a certain way that it is misleading and the quote kind of like a seminal piece in. In this topic I'm going to quote Darrell Huff who wrote how to live statistics. Back in 1954 statistical methods in statistical terms are necessary and reporting the mass data of social and economic trends business conditions. Opinion polls the census, but without writers who use the words with honesty, understanding and readers who know what they mean. There is so can only be semantic nonsense. So, the numbers by themselves can't really tell you a story that can't really tell you a certain argument or persuade you in a certain way it's only writers who are able to ethically use that data and convey.
[23:11] maybe a certain narrative, or. It also takes readers who may be able to like kind of understand able to interpret that data in order to make sense of it and maybe be exercise their judgment in terms of saying like this is an appropriate use of data or this is ina ppropria te use or some, someone is like not being forthright with how they're kind of conveying information. And so I'm going to kind of go through a couple of different examples of how data can be misused in this way, and how it can be used to mislead.
[23:50] And so the first example would be something called sampling bias. And so this is a statistical practice where someone is trying to essentially it's a way of selecting like a small group of individuals or estimates certain characteristics about a population so there are claims that are being made. When a sample is more stronger claims are made when a sample is more representative of a whole rather than taking like a small section of those individuals, and so that sampling bias occurs when members of an intended population are over represented or under represented it so it also occurs when the people who are gathering the data or compromised in some way. So, for example, like, I think a common example
[24:54] of sampling bias is like a slogan that you see in a lot of, like, dental products like nine out of 10 dentists recommend using Colgate using this certain toothpaste. And there are like certain assumptions that are being made there in that like, Oh, that must mean that like hundreds of our tens of thousands of Dentists have been surveyed on this. And then like all like all these dentist dentist definitely like responded to that survey and they definitely don't have an incentive to lie to everyone if they were doing this on their own volition.
[25:36] But so there's several assumptions that are being made, when you come across a statement like this so who was being surveyed what was the sampling size so it could be the case that only like 50 dentists were surveyed. And so that's definitely not as representative of like the entire dentistry profession, and who are those particular respondents are they actually dentists to where they forced to kind of like show their credentials or were, was it just lik e people that were kind of like found in the straight and saying like, oh, like, You are a Dentist professional. And are there like incentives to lie were people that were surveyed. Did they were they, for example, paid to give their opinion, or is this like a particular.
[26:31] Do they. How should I frame this. It doesn't benefit their career in any way to like respond to a survey, maybe if it's from a very prestigious outlet. So that could be some examples of why you might want to think about, are there any incentives for people to not be forthright. And I think this also plays into when people are asked about very sensitive topics like topics about money, or about drug use or sex. These are topics that people tend to either exaggerate or kind of conceal their behavior or activities.
[27:17] So some examples of sampling bias might be like an old, like an alumni report that says like the average, University of Alabama graduate and makes over $100,000 per year. So it's worth asking like how many like alarms were surveyed was this only like business students was this only like engineering students, how many like total Alabama graduates were surveyed. And then another example being like a new poll states at 85% of Americans have never consumed an illegal substance.
[27:55] People are incentivized to lie about those topics because they might not want to admit to doing anything illegal in a survey they might not trust the survey taker. So it's worth kind of considering like how might the sample that was gathered from a particular survey might be biased in a particular direction. Another method that can be used another statistical method that people use to maybe mislead others is cherry picking and so cherry picking is the practice of only using a small subset of data to kind of sell a certain narrative. So I think I brought up, Apple as example earlier of maybe a company that is only. They might only want to share a particular subset of data, kind of like make themselves look good and maybe try but drive up market share of of their stock and saying like, like this.
[29:02] This particular quarter like this technology product like sold like this many units, but it doesn't look at like the years long trends of like that particular product not selling well and actually declining and sales. And so debates over issues like climate change and crime rates are a big victim of cherry picking because rapid changes from year to year, do not account for years or decades on trends.
[29:33] So, a lot of people who engage in like climate change denialism, kind of like show like maybe a small subset of climate data saying like, Oh, it's actually showing that the earth is cooling over time or there's just like so much lik e rapid changes that you can't really say that the earth is like warming. But then it's not taking into account like decades long data that's showing that the amount of the temperature of the Earth is actually rising at and unsteady rate, and it's going above like certain industrial levels. And the same is true for crime data as well.
[30:14] This is a screenshot from the San Francisco, San Francisco Chronicle. They had a tweet on a few months ago saying that car break ins are up, 753%, and that like begs the question of like what what actually happened in. In 2020, there was like a massive like pandemic that caused a lot of people to stay inside. And so that is an accounting for maybe like us like maybe five years of data or 10 years of data that could show that like crime.
[30:47] Could be actually decreasing there's a lot of data to suggest that crime is still steadily declining across the years and especially over the decades since the 1970 is another statistical. I guess method that people use to mislead is to conflate correlation and causation. Just because I'm sure you both have heard this before, just because something is correlated, or two things are correlated doesn't mean that one is the cause of the other. So for, even if like a claim, turns out to be true like in this example. So, just to kind of describe this graph here, it's showing that countries that didn't implement a mask mandate right away, had a huge explosion in cases. Well, that countries that did mask up had fewer cases.
[31:50] This is an example of something that two things that are correlated. There could be a whole bunch of different factors that might explain this correlated phenomenon. A lot of those countries, for example, South Korea, Japan, Singapore, these are all like East Asian countries. Maybe it doesn't document. Maybe particular practices in those countries that don't have to deal with masking. It could be the case that the virus was not as prevalent in those communities where, whereas it was more prevalent in these European and North American countries.
[32:30] There are so many other different factors that could explain that, even if it turns out that mass are very helpful in curbing the spread of coronavirus. It's important to kind of like be kind of aware of that and be kind of skeptical about arguments that say that this like one singular issue is the cause of this very complex problem. So it is worth kind of like questioning, something, even if it turns out that, that kind of thing might be might be true in certain cases and mass are very helpful and carbon described but it doesn't account for kind of the multifaceted complexity of why coronavirus cases are rising in some places and not others.
[33:22] And then there are ways that visualizations are used to mislead people, and there are certain practices that are used within visualizations to mislead people. So one example of this is trunk hating graphs and omitting the baseline on the y axis. So, in both of these examples, the baseline is at this very incredibly high number so like in this example, 94 million is like the cutoff point when it really should be zero. And so the effect of that is that it can misrepresent data as more exaggerated or more dramatic than it actually is. And, yeah, so this chart on the right, it says that over 100 million people now receiving federal welfare.
[34:16] And it's kind of like looking at this kind of like explosion of cases over time from 2009 to 2011. But if we had the baseline at zero you could see that that's like a very marginal change compared to the actual numbers. And I think that chart counts anyone receiving federal welfare as anyone residing in a household and, which at least one person received a program benefits that could be Medicare, Medicaid, or, sorry so anything. Here's countenance tested welfare not Social Security or Medicare. Okay, so it could be like anyone in a particular household that's affected so that could be Medicaid or a different federal assistance program like.
[35:06] Temporary Assistance for Needy Families, for example. So anyone who is receiving that benefit. They count everyone in the household as receiving that benefit so that's just something to keep in mind for that particular chart. And the one on the left. There are many problems with this chart, but one is It's kind of exaggerating the, the shortness of women from South Africa, India, because the baseline is set at very close to five feet and so this isn't like equal intervals.
[35:47] So it kind of like makes it seem like people from are women from Latvia and Australia are gigantic compared to women from South Africa and India when it just is just a couple inches, and not like a huge like marginal or significant difference. Another example is just more manipulations of the white access. So intervals between x and y axis should be even consistent and visualizations that manipulate those intervals, do so by exaggerating, an increase or decrease through the use of like labels or illustrated elements for color. And so, like this example on the left, if y'all can see it is cancer screenings, and preventive services are going down for Planned Parenthood but abortions are going up from 2008 to 2013.
[36:52] But the issue is that these two numbers should be like completely different. This is like going from like 2 million to 900,000, but for some exam. For some reason the 300,000 abortions or, or like above this number right here so the numbers don't even seem to be on the same plane there it's like a manipulation of the white access. Whereas for this example on the right under President Obama more students are earning their high school diplomas than ever before. Even though it is like a marginal increase from 75% in 2009. To 82% into 2015.
[37:38] It's kind of like shown that this is like over like half of like this particular number based on the amount of books that are being stacked. And so, like, since the baseline. Since the y axis like being manipulated here. You can't like really see how profound the difference actually is for the number of students with high school diploma.
[38:09] I would say one of the most notorious examples of going against. Kind of like manipulating people or misleading people with data visualizations is going against conventions going against standard conventions of data visualizations like the why access should start at the bottom left and should start at zero. That's like a pretty common convention. When it comes to line graphs and bar graphs.
[38:44] And the reason for that is that charts and graphs are have become more standardized and to go against that standardization against those conventions, can be very confusing for people at first glance, and so it's kind of generally agreed upon that you should like start the y-axis at zero at the bottom left so you're not confusing people. And this kind of like notorious visualization starts the y axis I'd like 1000, to convey an argument that after 2005.
[39:23] After the Stand Your Ground law, the number of murders committed using firearms actually decrease. But the problem is, is that if you actually flip this to where it should be zero was actually down here. You could see the number of murders committed using firearms actually increased after that law was implemented in Florida. And so, that's like one example of just like how going against conventions can mislead people, because it's not properly.
[39:55] Starting the y-axis at its proper point. One is, this example right here is actually going with conventions, what population density maps. Usually darker colors are meant to represent that something is more densely populated so the states of California, Illinois and New York Massachusetts, New Jersey are more densely populated than Wyoming and Montana. But if that convention was reversed in some way. If people if, for example you tried to say that the lightly shaded colors indicate more densely populated place.
[40:41] That would confuse readers who are already used to knowing that a place that is lightly shaded should be less densely populated so they might confuse certain places to be more popular than then they should be. And so that that can be pretty important when you see like a visualization like on social media like on Twitter or Instagram, and you're just kind of like briefly glancing on it, you're not really studying it in depth. If it goes against a certain convention that could easily like mislead someone to thinking, a certain idea is true or not. And then, I'm using the wrong chart, I can be a pretty big way of misleading people or at least confusing them pie charts for example are notoriously misused. They're supposed to represent parts of a whole.
[41:34] So this example on the left doesn't really make sense of Americans who have tried marijuana 51% say today 43% say last year 34% say 1997. Yeah, that just doesn't make any sense because if you're trying to gauge like how maybe marijuana use has changed over time you would use a very different kind of visualization to display that versus like a pie chart which should always like, add up to 100% and it should be kind of like parts of a whole it should be like the show like how the variable represents, or how it's related to the variable and how it's representing parts of a whole versus certain other types of visualizations. There are data that correspond with certain types of visualization so it's data that is being used to compare two values to the differences
[42:40] or similarities and those values who would use comparison so like bar charts dumbbell plots are pretty good for those who are trying to kind of gauge relationships that heat map. For example, you're trying to show, whether it's connections and data or depicting a correlation or lack thereof of heat maps or scatter plots are good for that distribution visually visualizations that display frequency or maybe changes over time, histogram would be a lot better for this kind of data right here this kind of pull showing a change over time compared to a pie chart.
[43:20] Alright so now we are going to rank visualizations. And so, since we have to have you here I'm going to put two PDFs in the chat. Feel free to maybe choose one PDF over the other, but essentially you'll be ranking visualizations from which visualization is most reliable vs. which ones are the least reliable, and just using what you've kind of like learned today about how visualizations are represented.
[44:00] How people maybe are misusing certain visualizations to convey an argument. You can see just like what visualizations I'd be more reliable versus what might be least reliable. So I'm going to stop. Now what I'm going to do is actually put this in the chat so everyone can see it. Let's see if I can actually share a PDF here.
[44:48] one second here. I think what I'm going to do is, I don't think I'm actually able to share files in the chat. My apologies. So what I'm going to do is gonna do a screen share of one of one of those files that contains all those visualizations that you all can see that file and be able to kind of comment which one is most reliable versus which one is least reliable.
[45:19] So let me go ahead and find that, and then I will share that everyone. What I'm going to use second one. Alright so I'm going to share my screen one more time. Okay.
[46:08] And you all can see this PDF right now. Give me a thumbs up if you can. Awesome. Yeah, great. So, um, let's see these four visualizations right here I'm going to zoom out of it. Okay, let's see.
[46:38] Okay, so we have two visualizations right here one is a population density map, one is kind of a donut or another kind of like pie chart. That is showing how many seats that the Labour Party in the UK one versus how many seats of the Tories one based on your first impressions of these visualizations. How would you characterize their reliability. One seems more reliable than to definitely.
[47:22] It's definitely following the conventions of a population density maps of red is used to indicate that places more as more population density versus colors like yellow and green, and then yes the one from The Sun. I think one thing to kind of like note is that the UKIP party is kind of like, even though it didn't win any seats is shown to like have at least half of what the Liberal Democrats won with that with this chart, and the SNP almost has like the exact same area as the Liberal Democrats, even though they won more.
[48:11] So it is definitely misleading like how this is represented here is even though this party didn't even win anything it's somehow like represented on this donut chart. Yeah, UKIP shouldn't even be showing it's at zero here exactly. All right, and then like these two other visualizations right here. One is showcasing types of debt by the average us household. And then this second chart at the bottom here.
[48:45] It's showing our spent online by age and gender. Let's using a bubble chart. One since more reliable than to hear. Yeah, I would say, yeah, the second one this is like a weird way to kind of like distinguish differences between like men and women by like showing. Yeah, it's hard to kind of like even discern what these different bubbles me even if it's like comparing something to something else.
[49:25] Yeah, so maybe 12 year old men are spending more hour, or spending less hours online than 12 year old women, it can be hard to kind of gauge that at first. Yeah, I think the different size of the circles is that supposed to convey that that those. This is this supposed to indicate that more people are spending time online there yeah it's it's kind of difficult to kind of gauge like how that is even relevant at all.
[50:04] Yeah, the bubble chart is quite unclear. Yeah bubble charts are often used to kind of demonstrate frequency. I think like heat maps for example, show like. If a place is like more densely populated or more people are responding to a particular issue like the bubble like expands it gets bigger. But in this context doesn't really make sense.
[50:37] You would think that like maybe a bar chart would be more appropriate there, whereas this one. Yeah, I would say it's this is probably the best way to capture that data is with comic the stacked bar there. It is like a little bit misleading though and that auto loans are taken to be like three fourths of mortgages Even though 28,000 is definitely a lot less than 176,000.
[51:12] And so I think that this chart here needs to be adjusted in terms of like how how much credit cards and auto loans kind of need to be like kind of cut in half shrunken significantly compared to mortgage debt and I and student loans especially. Alright so we're gonna go back to slides here and just kind of ask some final thoughts about how do you think you will engage with data in the future. Now that you've kind of like seen these examples of how data can be used to mislead knowing the importance of the context
[52:18] of data how it's gathered, how it's being represented and presented to give a particular argument whether that is for commercial rhetorical or political purposes. Yeah, I just, I think being able to be literate with data means being able to critically assess it and being able to know for example. Why data is being presented in a particular way and for what purpose. So I'm hoping after this presentation, you'll have the takeaway of. You can properly.
[52:55] See, data, whether that's, like, it's represented through social media or other channels like in scholarly sources for example and be able to kind of maybe look at the methodology section and see like how did they really gather this data can we properly trust. The argument that's been conveyed to me. So, yeah, I hope that folks kind of like how that takeaway after viewing this presentation today. But I'd be happy to answer any questions that you have. Let me just share my work cited page right here and just kind of sharing how I put this together today but I'm happy to answer any questions. Okay.
[53:46] In the chat here...I'm going to be careful pay attention to what information is intended to be conveyed also pay attention to things like cherry picking manipulation of axes. Yeah, absolutely. Will the slides be made available? Yes! The slides including my notes on those slides I will email to each of the registered participants And this recording of this presentation will also be made available as well All right! Wonderful! Thank you everyone for attending!
[54:29] And if you're watching this recording later, feel free to email me. I'll put up my contact information so that everyone can see that You can reach me via email at rtpeterson1@ua.edu or contact me via phone as well. I'm happy to answer your questions in whatever way you want to get ahold of me. So, thanks everyone! I really appreciate you reaching out and listening to my presentation today. Thank you!
在Vidleaf工作台中打开
搜索转录文本，选择行，复制带时间戳的引文，翻译。
署名
“这张图表在骗你！”——数据素养入门，由阿拉巴马大学图书馆（https://www.youtube.com/@universityofalabamalibraries）制作，采用CC BY 3.0（https://creativecommons.org/licenses/by/3.0/）许可。源视频：https://www.youtube.com/watch?v=0SDYnwa4ir4。本页面是视频的文本转录，添加了段落分隔和时间戳；创作者与Vidleaf无关，也不为其背书。Vidleaf添加了AI标点和AI生成的摘要；原始措辞仍可在本页面上找到。
您是创作者或权利持有人吗？请求更正或移除：copyright@vidleaf.app（参见关于这些页面）。
最后更新于