多个样本率的卡方检验及两两比较

《多个样本率的卡方检验及两两比较》由会员分享，可在线阅读，更多相关《多个样本率的卡方检验及两两比较（24页珍藏版）》请在装配图网上搜索。

1、SPSS ：多个样本率的卡方检验及两两比 较来自：医咖会医咖会之前推送过“两个率的比较（卡方检验）及Fisher精确检验的SPSS教程”，小伙 伴们都掌握了吗？如果不止两个分组，又该如何进行卡方检验以及之后的两两比较呢？来 看详细教程吧！1、问题与数据某医生拟探讨药物以外的其他方法是否可降低患者的胆固醇浓度，如增强体育锻炼、 减少体重及改善饮食习惯等。该医生招募了150 位高胆固醇、生活习惯差的受试者，并将其随机分成3组。其中一 组给予降胆固醇药物，一组给予饮食干预，另一组给予运动干预。经过6 个月的试验后， 该医生重新测量受试者的胆固醇浓度，分为高和正常两类。该医生收集了受试者接受的干预方法

2、（in terven tiOn和试验结束时胆固醇的风险程度 （risk_lev）l等变量信息，并按照分类汇总整理，部分数据如下：Total count data (frequencies)Individual scores for 色ach participant$ ch:-cquar&-:=Et-cf-homogeneity-lcE-sav |Datat=tI - IBM SPSS Statisti百? chi-tquar&-tEt-cfhcitTioganeity-3ioindividual-sct密0154 iist_lBVElintg mentionriskjGel1口 nigHigh

3、2rugHigh3DrugHigh斗nJ9High5DrugHighEDrugHigh7DrugHighBDrugHigh9nigHigh10DrugHigh11DrugiHigh12High13DrugHighuDrugHighISDrugHigh16DrugHigh17nigN omnsl18DruyNormal19DrugiNormal20DrugNoimaiTransform Analyzeinterventionriskjeveifreq1DrugHigh162nJ9Normal343DietHigh2a4OieiN Dririal225EjcerciseHigh30GExercis

4、eNormal2Q7注释：本研究将胆固醇浓度分为“高”和“正常”两类，只是为了分析的方便，并不代表临 床诊断结果。2、对问题的分析研究者想判断干预后多个分组情况的不同。如本研究中经过降胆固醇药物、饮食和运 动干预后，比较各组胆固醇浓度的变化情况。针对这种情况，我们建议使用卡方检验 （2C），但需要先满足5项假设：假设 1：观测变量是二分类变量，如本研究中试验结束时胆固醇的风险程度变量是二 分类变量。假设2：存在多个分组（2 个），如本研究有3 个不同的干预组。假设 3：具有相互独立的观测值，如本研究中各位受试者的信息都是独立的，不会相 互干扰。假设4:研究设计必须满足：（a）样本具有代表性，如

5、本研究在高胆固醇、生活习惯差 的人群中随机抽取150位受试者；（b）目的分组，可以是前瞻性的，也可以是回顾性的， 如本研究中将受试者随机分成3 组，分别给予降胆固醇药物、饮食和运动干预。假设5：样本量足够大，最小的样本量要求为分析中的任一预测频数大于5。经分析，本研究数据符合假设1-4,那么应该如何检验假设5，并进行卡方检验（2乂） 呢？3、思维导图、研究设计检齋數抵足舌满足假设1 f二分类变嚴)、假设2梦个仆 坦)、假设d (独立性和假设4猗11样和分an一、SPSSJK 作在Crusstab顶下棉在数祐启舌满足愷说5 样本址)三、决定四、决定进疔Fi曲er精确检验報牺预测频辿利断：样木駄是

6、否足堰是否四，决定魅行卡方检验(2C)4、SPSS 操作4.1 数据加权在进行正式操作之前，我们需要先对数据加权，如下(1)在主页面点击 DataWeight CasesDataSstl - IBM SPSS StatisticsGraphsAdd-oSort Cases.Sort Variables.& Aggregate.Rake Weights.Propensity Score Matching.Case Control Matching.Split into Files聯 WeighiCsses.Identity Duplicate Cases.Comar已 Datasets.Copy

7、 Dataset1 .IO1inters1D2D3C4C5Exe&Exe7&扌10111213U1516171819202122弹出下图： Define Variable Properties.Copy Data Propemes.Define Dates.Define Multiple Response Sets.国 Transpose-Re structure.Split File.(2)点击 Weight cases by, 激活 Frequency Variable窗 口(3)将 freq变量放入 Frequency Variable栏(4)点击 OK4.2 检验假设 5数据加权之后，

8、我们要判断研究数据是否满足样本量要求，如下：(1)在主页面点击 AnalyzeDescriptive StatisticsCrosstabs-scoret.sav DataSetl - BM SPSS Statistics Data EdiLoiTransform Analyze Graphs UlilitiesAdd-o nsWindow He：pin1erientionrisk level1DrugHigh2DrugHigh3DrugTigh4DrugHigh5DrugHigh6DrugHigh1DrugHigh8DrugHigh9DrugHigh10rugT妙11rugHigh12Dru

9、gHigh13DrugTigh14rugT妙1SDrugHigh16DrugHigh17DrugTighReportskDescriptive StatisticsCompare Means卜General Linear kcodelGeneralized Linear ModelsMixed ModelsCorrelate卜RegressionLogli卜ClassityDimension RedccticnScaleNon parametric TsstsForecastirigSurvival卜Multiple Response朝 Simulation.Quality ControlEH

10、 ROC Curve.Frequencies.-DeecriptivAS囲 P-P Plots.Q Q Plots.A Explore.0 Crosstabs.恰TURF AnalysisJ Ratio.弹出下图：st c畫Erba.-I口 LSD-Rye-usi(D(DCIL2dla.rtg5匚吕BSSll-hRb-BS刀lloMlE-lloQ-Llmn(3KB- La.yn)2prhglaus - zest一 一 口亦 pFy -6茜曾會 inErb-eEFy 豊He-P(2)am*inr+ervenr+io営 risrlevea-n_J Row(s)族啓 column(s)族口 口 is

11、:p-lljy Qu gi(D(DdIffaJ cnQjrtg1 Bn 吕fDww酉五ul K Eamre 囚5黑 Ganne- He-D(3)点击Statistics弹出下图:博匚rosstabs; StatisticsX._i Chi-square CorrelationsMomiinalrdiinal Contingency co efficient Gamma.Phi and Cramer*s V.SomersdB Lamtidaii Kendalls tau-bUncertainty coefficientj Kendalls tau-cNominal by Interval二.Ka

12、ppaB Eta Ri3k曲:Nemar Cochrans and Mantel-HaenszelTest eommcn odds ratio equals；:continue Gajicel Help (4) 点击 Chi-square匚rosstab5： StatisticsH Chi-squarei CorrelationsrOrdinalrNominal Contingency co efficientPhi and Cramers VLambdaUncertainty-co efficient Gamma Somers dEKendalls tau-b.j Kendlalls tau

13、-crNominal by IntervalKappaEtaRiskMcNemarCochrans and Mantel-Haenszel statisticsTest common cdds ratio equals: -iCoritiriLig C日口饲IH创p (5)点击 ContinueCells醴 Civstabs Cdl Display-Counts-Zr-test回迫 bserved Compare column proportionsl 二 ExpectedO Adjust p-values (Bonferroni method)Hide small countsLes tha

14、nr tsi 11 Ldy t bKE! b lULld lbO RawUn standardized Calumn StandardizedE Total Adjusted standardizedNoninteger Weights Round cell counts Round case weights Truncate cell counts Truncate cas weights No adjustments(6) 点击 Counts 栏中的 Expected 选项也匚rosftabs: Cell DisplayX-CountsZrtest Observed Cornare col

15、umn proportionsHExp-ected| Adjust p-values (Banferroni method)Hide small countsLess than h-Noninteger Weights Round cell counts O Round case weightsQ Truncate cell counts O Truncate case weightsQ No adjustmentsContinue Gan cel Help(7)点击 ContinueOK经上述操作，SPSS输出预期频数结果如下:Level of cholesterol risk: High

16、& Normar1* Type of intervention: DrugDiet & Exercise11 CrosstabulationExpected CountType of intervention: Drug, Diet& ExerciseDrugDietExe rciseTotalLevel of chclesterol risk:High24724.724.74.0HighMo rm als 8sNormal25.325.326.376.0Total50.050.050.0150.0该表显示，本研究最小的预测频数是24.7，大于 5，满足假设5，具有足够的样本量。Chi-Squ

17、are Tests表格也对该结果做出提示，如下标注部分：Chi-Square TestsValuedfAsymptotic Significance (2-sided)Pearson Chi-SquareB.1752.010Likelihood Ratio9.2362.009Li near-by-Lin ear Association7.7891.005N of Valid Cases150a. cells (0.0%) have expected count lessthan 5. minimum sxpected count is 24.67.The即在本研究中，没有小于5的预测频数，可以

18、直接进行卡方检验(2C)。那么，如果 存在预测频数小于 5 的情况，我们应该怎么办呢？一般来说，如果预测频数小于5，就需 要进行Fisher精确检验(2)，我们将在后面推送的内容中向大家详细介绍。4.3卡方检验(2)的SPSS操作(1)却 test-f-proptHio ns-i n-sps-s- in di vi du-a I- sc ones, wv Dei a 5d:1 - IBM 5PS5 Statistics Onto Edit”File Edit 里悄別 Cata Transrorm .Analyze空gj BBJ.产3B. interunllon|1intervientionns

19、le/BlJL1DrugHigh2DrugHigh3DrugHigh4DrugHigh5DruHighEDrugHigh7OrUgHigh8DrugHigh9DrugHigh10DrugHigh11DrugHigh12DrugHigh13CruHigh14DrugHigh15QrUgHigh16DrugHigh17DrugHighRetorts卜DpcripW& SlaHstcsCnrnpare MeanskGeneral Linear ModelG ph 9railed Lin ear M odi srdModelsCorrelate卜RegressionLoglinarCl3EEitC)的

20、结果分析之前，我们需要先对研究数据有个基本的了解。SPSS 输出结果如下：riskjevel * intervention Crosstabultionirte rve ntioriTotalDiuqDietExerciseriskjevel HighCountExpected Countft within inteiventiori1弘28b30b7424.732.0%24.756.0%24.760.0%74.049.3%Norriial CountExpected Count% within intervention34a22t20b7625.368.0%25.344.0%25.340.0

21、%76.0507%TotalCountExpected Count% within intervention5050.0100.0%5050.0100.0%5050.0100.0%150150.0100.0%Each subscript letter denotes a subset of inteiventiori categories whose column proportions do not differ significantly from each other at the .05 level.该表提示，本研究共有150 位受试者，根据干预方式均分为3 组。在试验结束时，药 物干

22、预组的50 位受试者中有16 位胆固醇浓度高，饮食干预组的50 位受试者中有28 位胆 固醇浓度高，而运动干预组的50 位受试者中有30 位胆固醇浓度高，如下标注部分：也* interueiition Crosstabulationinte iventionTotalDrugDietExerciserisk_l evelHighCount让a28b30b 74Expected Count24724.724.774.0% wiihiri intervention32.0%56.Q%60.0%49.3%NormalCount如22b20b76Expected Count25.325.325.376

23、.0% within intervention68.0%44.0%40.0%50.796TotalCount505050150Expected Count50.050.050.0150.0% within intervention100.0%100.0%100.0%100.0%Each subscript letter denotes 爼 subset of intervention categories whose column propoiiions do not differ significantly from each other atihe .05 level.由此可见，药物干预比

24、饮食或运动干预的疗效更好。同时，该表也提示，药物干预组 的50 位受试者中有34 位胆固醇浓度下降，饮食干预组的50 位受试者中有22 位胆固醇浓 度下降，而运动干预组的50 位受试者中只有20 位胆固醇浓度下降，如下标注部分：riskjevel * intervention Cro-sstabulationinten/entionTotalDrugDietExe rciseriskjevelHighGaunt16s28b30l74Expected Count24.724724.774.0% within intervention32.0%56.0%60.0%49.3%NormalGaunt如

25、22b20b |Expected Count25.325.325.376.0% within iriteivention68.0%44.0%4O.a9f50.7%TotalGaunt505050150Expected Count50.050.050.0150.0% within interventiori100.0%100.0%100.0%100.036Each subscript letter denotes a subset of iriteivertion categories whose column propoitions do not differ signifieantlfrom

26、 each other at the .05 level.但是，当各组样本量不同时，频数会误导人们对数据的理解。因此，我们推荐使用频 率来分析结果，如下标注部分：riskjevel * intervention CrosstabulatianinterventionTotalDrugDietExerciseriskjevelHighCount16a28b30b74Expected Count24.724.724.774.0% within inteivention32.0%56.0%60.0%49.3%NormalCount34a22b20b76Expected Count25.325.325

27、.376.0% within intervention68.0%44.0%40.0% |50.7%TotalCount505050150Expected Count50.050.050.0150.0% within intervention100.0%100.0%100.0%100.0%Each Eubscript letter denotes a sublet ofinteivention categories whose column prapoitions do not differ significantly from each other atthe .05 level.该表提示，药

28、物干预组的 50 位受试者中 68% 胆固醇浓度下降，饮食干预组的 50 位受 试者中44% 胆固醇浓度下降，而运动干预组的50位受试者中只有40% 胆固醇浓度下降， 提示药物干预比饮食和运动干预更有效。但是这种直接的数据比较可能受到抽样误差的影 响，可信性不强，我们还需要进行统计学检验。5.2卡方检验（2 乂）结果本研究中任一预测频数均大于5，所以根据Chi-Square Tests表格分析各组的差别。SPSS 输出检验结果如下：Chi-Square TestsValuedfAsymp. Sig,(2-sided)Pearsori Chi-Square9.175a2.0-10Likeliho

29、od Ratio9.3362.009Linear-by-Linear Association7.7S91.005N of Valid Cases150a. 0 cells (0.0%) have expected countless than 5. The minimum expected count is 24.67.卡方检验（2C）结果显示x2=9.175 P = 0.010说明本研究中各组之间率的差值与0 的差异具有统计学意义，提示药物干预与饮食、运动干预在降低受试者胆固醇浓度的作用 上存在不同。如果P0.05，那么就说明各组之间率的差值与0的差异没有统计学意义，即 不认为各组之间存在差

30、异。5.3卡方检验（2乂）中的成对比较分析如果卡方检验(2C)的P0.05，说明至少有两组之间的差异存在统计学意义。SPSS 输出的risk_level * int erven tion Cross tabu表at格通过数字标记提示了两两比较的结 果，如下标注部分：* intervention CrosstabulationinterventionTotalDrugDietExerciserisk_ levelHighCount2眺74Expected Count24724.724774.0% within intervention32.0%56.0%60.0%45.3%NormalCount

31、34s22h20t.76Expected Count25.325 325.376.0% within intervention68.0%44.0%40.0%50.7%TotalCount505050150Expected Count50.060.050.0150.0% within intervention100.0%100.0%100.0%100.0%Eath Gubscript letter denotes s subset of iriteivention categories whose column proportions do not differ significantly fr

32、om each other at the .05 level.大家可能会注意到，每组数据的标记相同(即上下两行的标记相同)，那么我们只要 知道组间标记的作用即可。怡* irrterwiTtion Crosslabulationinte ive ntionTotalDrugDietExerciseriskj 已 y已 IHighCount28b30|74Expected Count讨24424十74.0% within intervention32.049.3%NormalCount34a20b76Expected Count25 325.325.376.0% within interventi

33、on68.0%44.0%40.0%50.7%TotalCount50SO50150Expected Count50.050.050.0150.0% within intervention100.0%100.0%100.0%100.0%Each subscri pt letter denotes a subset of i rite mention categories whose column propoitions do net differ significantly from each other at the .05 level.那么，risk_level * in terven ti

34、on Cross t abula格的标记是什么意思呢？第一种情 况，各组间无差异，如下：11Number of Independent q旧tmG4up1Group2Gm up3Gmup4SraMp 号GiiJp61Qrou ps 1 tc 3 己rs / dtfferent： Groups 1 and 2 fa1 versus 閃卜 Groups 1 nd 3 versus sc/h Bnd Groups 2 and 3 fbb versus sabc2I he following jm irwlae com parisons a re staiislicaiiy SJniiicani di

35、fferent： Graups i and 2ver&us Bb*| Groups 1 and 3 versus bes Groups 1 and 4 (a* versus d*, Groups 2ad3 fb versus 昭h Gmups 2 日nd 4 fb ver&us dT and Groups 3 and 4 W venuiabcd3Tha following psirwl&e compariEDns 日 re 血诞 Licalty gnrficant dtfferwit. Groups 1 and 2 fs8 versus b,. Groups 1 And 3 (eT versu

36、s Bea|ffl Groups 1 and 4 (a versus dX Gr&ups 1 and 5扩 versus el专荀ups 2 日nd 3 pbB versus gTl Groups 2 mricM jb8 versus bcT). Groups 2 and 5 fba versus e, Gnsup 3 and 4 (c1* versus d1 and Groups A and 5 (d11 veraus Pj日bGde即每组标记字母均不相同，说明任意两组之间的差异均存在统计学意义。第三种情况，有些组之间存在差异，而另一些组之间的差异没有统计学意义，如下：Number of I

37、naependerrt gr-GtiosPairwiseGroup1(jrDtfp2irDup I3 11There is a stalistkalJy signiHcant drfrigrflnce- in propcnions between Group 1 and Group 2 r*F versus 七ad Group 1 end G roup 3verEub but there is no sbatrstJcally significantly tf-erenee In proportions betvreen Groups 2 end 3 fb1* versus b-.9bb2Th

38、ere* 色 a 如町玄tic对妙 signirani: dfflFeierice pfopaiibris betwean Group 1 and Grcbup 3 (BaB versus b1), but there c no HnUFliGal忖 signAlTGaintly diFfcrcncc in proportions beteen Groups 1 and 2 (,iaver54fs al and Groups 2 snd 3 fb versus b8)苜4b3There Is e E-tallstkzally significant difference tn prop oni

39、ons betiwen Groups 1 ard 2 fe* vrj&and Groups 2 and 3rtrsuibut諒 no stalislically sJgnifkiiBrvtMcjiffcr&ncc irti 戸怕曲沁 出帕殆MU芮 1 flrtd 3 YCrSuS 91axb如果任两组之间标记字母相同，说明这两组之间的差异没有统计学意义；如果两组标 记字母不同，说明这两组之间的差异存在统计学意义。根据这一原则，分析本研究结果如下：* int&rveiitio-n Crosstabulati-oninterventionTotalDrugDietExerciserisk_ lev

40、elHighCount28b30|74Expected Count2+724.72474.0% within intervention32.0%56.0%60.0%49.3%NormalCount34a22b20t76Expected Count25.325 325.376.0% within intervention68.0%44.0%40.0%50.7%TotalCount505050150Expected Count50.050.05Q.Q150.0% within intervention100.0%100.0%100.0%100.0%Each subscript letter den

41、otes a subset of i rite mention categories whose column proportions do not differ significantly from each other at the .05 level.该表说明，在本研究中，药物干预的降胆固醇作用(“a”)与饮食干预的降胆固醇作用 (“b”)的差异存在统计学意义(P0.05),药物干预的降胆固醇作用(“a”)也与运动干预的降胆 固醇作用(“b”)的差异存在统计学意义(P0.05)，而饮食干预(“b”)与运动干预(“b”)在降胆固醇 的作用上没有差异。6、撰写结论6.1若卡方检验(2C)的P

42、0.05本研究招募150 位高胆固醇、生活习惯差的受试者，随机分组后分别给予药物、饮食 和运动干预。试验结束时，药物干预组有34 位(68%) 胆固醇浓度下降，饮食干预组有22 位(44%) 胆固醇浓度下降，而运动干预组有20 位(40%) 胆固醇浓度下降，三组差异具有统 计学意义(P=0.010)。成对比较结果提示，药物干预的降胆固醇效果好于饮食或运动干预(P0.05),而饮食 与运动干预在降低胆固醇浓度上的作用无差异(P0.05)。6.2若卡方检验(2C)的P0.05本研究招募150 位高胆固醇、生活习惯差的受试者，随机分组后分别给予药物、饮食 和运动干预。试验结束时，药物干预组有24位(48%)胆固醇浓度下降，饮食干预组有22 位(44%)胆固醇浓度下降，而运动干预组有20位(40%)胆固醇浓度下降，三组结果的差异 没有统计学意义(P=0.620 )。