多個(gè)樣本率地卡方檢驗(yàn)及兩兩比較之spss超簡(jiǎn)單

1、SPSS多個(gè)樣本率的卡方檢驗(yàn)及兩兩比較來(lái)自：醫(yī)咖會(huì)醫(yī)咖會(huì)之前推送過(guò)“兩個(gè)率的比較(卡方檢驗(yàn))及Fisher精確檢驗(yàn)的SPSS教程”，小伙 伴們都掌握了嗎？如果不止兩個(gè)分組，又該如何進(jìn)行卡方檢驗(yàn)以及之后的兩兩比較呢？來(lái)看詳細(xì)教程吧！1、問(wèn)題與數(shù)據(jù)如增強(qiáng)體育鍛煉、減某醫(yī)生擬探討藥物以外的其他方法是否可降低患者的膽固醇濃度, 少體重及改善飲食習(xí)慣等。該醫(yī)生招募了 150位高膽固醇、生活習(xí)慣差的受試者， 并將其隨機(jī)分成 3組。其中一組 給予降膽固醇藥物，一組給予飲食干預(yù)，另一組給予運(yùn)動(dòng)干預(yù)。經(jīng)過(guò) 6個(gè)月的試驗(yàn)后，該醫(yī) 生重新測(cè)量受試者的膽固醇濃度，分為高和正常兩類。該醫(yī)生收集了受試者接受的干預(yù)方法(i

2、ntervention)和試驗(yàn)結(jié)束時(shí)膽固醇的風(fēng)險(xiǎn)程度(risk_level )等變量信息，并按照分類匯總整理，部分?jǐn)?shù)據(jù)如下：注釋：本研究將膽固醇濃度分為“高”和“正?！眱深悾皇菫榱朔治龅姆奖?，并不代 表臨床診斷結(jié)果。2、對(duì)問(wèn)題的分析研究者想判斷干預(yù)后多個(gè)分組情況的不同。如本研究中經(jīng)過(guò)降膽固醇藥物、干預(yù)后，比較各組膽固醇濃度的變化情況。針對(duì)這種情況，我們建議使用卡方檢驗(yàn) 但需要先滿足5項(xiàng)假設(shè)：飲食和運(yùn)動(dòng)(2 X C),假設(shè)類變量。1 :觀測(cè)變量是二分類變量，如本研究中試驗(yàn)結(jié)束時(shí)膽固醇的風(fēng)險(xiǎn)程度變量是二分假設(shè)2:存在多個(gè)分組(>2個(gè))，如本研究有 3個(gè)不同的干預(yù)組。假設(shè)干擾。3 :具有相互

3、獨(dú)立的觀測(cè)值，如本研究中各位受試者的信息都是獨(dú)立的，不會(huì)相互4:研究設(shè)計(jì)必須滿足:假設(shè)的人群中隨機(jī)抽取150位受試者；本研究中將受試者隨機(jī)分成3組,(a) 樣本具有代表性，如本研究在高膽固醇、生活習(xí)慣差(b) 目的分組，可以是前瞻性的，也可以是回顧性的，如 分別給予降膽固醇藥物、飲食和運(yùn)動(dòng)干預(yù)。假設(shè)5:樣本量足夠大，最小的樣本量要求為分析中的任一預(yù)測(cè)頻數(shù)大于5。經(jīng)分析，本研究數(shù)據(jù)符合假設(shè)1-4，那么應(yīng)該如何檢驗(yàn)假設(shè)5,并進(jìn)行卡方檢驗(yàn)(2 X C)呢？3、思維導(dǎo)圖4、SPSS操作4.1數(shù)據(jù)加權(quán)在進(jìn)行正式操作之前，我們需要先對(duì)數(shù)據(jù)加權(quán)，如下: 在主頁(yè)面點(diǎn)擊 Data 7 Weight Cases彈

4、出下圖:Weight CasesIgjjj interventfcri di risfcjevei freq1 Ooi not wrieight casesO Weight M&es byFll '!二 I'-'3111：-* Current Status* Weight cafi&s by fr&qOKPaste 11 ResetCancelHelp(2) 點(diǎn)擊 Weight cases by ，激活 Frequency Variable窗口嗎 Weigiht CasesR intervention cjj TisKJ 列創(chuàng) 爐 freq

5、69;0 natweigtit cases.jIBMIBJM IIMJIM MUBJII |©(WeiglTl cases byjFrequency Va'rii3 QI e:Current Stains. Weigh! cases Dy fr&qR&set CancelHelp誑)intE“erition d riskjevelOK(3) 將 freq 變量放入 Frequency Variable 欄繪 Weight CasesO Do not術(shù)eight cases. Weight cases by Requancy Variabls- Current

6、Status: Weiont cases Dy ITeqif'1 (fPaste JJ Cane巳IHelp(4) 點(diǎn)擊OK 4.2檢驗(yàn)假設(shè)5數(shù)據(jù)加權(quán)之后，我們要判斷研究數(shù)據(jù)是否滿足樣本量要求，如下:T Crosstabs(1) 在主頁(yè)面點(diǎn)擊 Analyze f Descriptive Statistics彈出下圖:駅 Crosstabs Difiplav dusiered par charts Suppress tablesX匕Fj蘭 Reset I CancelHelp(2)將變量 intervention和 risk_level 分別放入 Row(s)欄和 Column(s)欄嗆

7、 Crosstabs-Layer 1 of 1 Display clustered bar charts Suppress ta&lesROMS')：Exact.XJ-isK_lev&lStatistics. 1CeIJs. 11JColumn(sJ<Form at. |intervention鼬底“ 1XC - lI-J； lj；r Jl -J |rj _-| _ _ f I T J - IOK J'PasteKe set Cancel* IHelp(3)點(diǎn)擊Statistics ，彈出下圖:Crosstabs: Statist!cs iChi-squa

8、re - n IT-nilin-iniii iii ! Correlationsr Nominalr Ordinaln Continig,ency coeUcienitI'【Phi and Cramer's V二 LambdaC Uncertainty coefFicient匚I GaTTiirnaLJ Somers' d Kendalls iau*b Kendall's tau<Nominal Dy Interval .Eta !<appa Risk McWernarI ' I Cghr即5 and Maitel-Haenszel st蘋t

9、i了tiosJIl iii_ - d I ,1-1-Continue Canc&l | H&lp(4)點(diǎn)擊 Chi-square© Crosstabs: Statist!C5 H H HB H HHiChi-square! r. nip inil or'i Correlationsr NominalrOrdifial Contingency coefficientLJ Phi and Cramers V Lamnda Uncertinbr coefficient!?_. Garnrnai' Somers' drKendalls lau*bC K

10、endall's tau-cNomiinal by Interval Eta Kappa RiskWcNemarL I Cochran's and Mantel-Haenszel statisticsIIillCentinue Cancel Help(5) 點(diǎn)擊 ContinueCells-Z'test'C ountsn Compare column proportions0 -C. IJ'I |A 孫三 II-H ：l I 1 11 111 l-ni lllll III IIFT1II閘也 tJSEFVEd H H . . . H XD peeled

11、Hide small countsLess than 匸P &rce ntagesrResiduals gciw Column lolal Unsta ndardi 軍匸I Standard'ilzed Adjusted standardizedNoninteger Wsi ght s Round cell counts O Round case weights O Truncate c&y 匚ounts O Truncale ca&e weights O No adjustmentsContinueCancelHElp(6)點(diǎn)擊Counts欄中的Expecte

12、d選項(xiàng)Crosstab!； Cell Dh,plfiiyCounts2-test ObservedIV lExpectedlj111 _ Illi UlBllliihlllBISIBIIIUII llUdlll UiiU IliaJ-_ Hide small counts Compare column poportionsH' 7匚.2-.,dL! IlC IJ II “ ：IC-Percentages一 Row ColumnTotal-F? esidual Un slandia rdized Standardized Adjusted stand3rdiz&<j-Non

13、integer Weights畫 Round cell counts O Round case weights O Truncate cell counts OTruncate case weights © No adjustrrients(£Qntiriu&j CancelHelp點(diǎn)擊 ContinueOK經(jīng)上述操作，SPSS輸出預(yù)期頻數(shù)結(jié)果如下:Level of cholesterol risk: "High" & "Normal*'* Type of intervention: "Drug'&qu

14、ot;,'Diet'* 5. "EKGrcisG" CrosstabulationEKpectedl CountType ofintervenlion: "Drug", “Di或& "ExerdsG"DrugDhtExerciseTolalL&vel of cticlest&rol risk:High34.724 724.774.0High" £.''Nornnal"Ncrnal25.325.326.376.0Total501050105 DIO15

15、0.0該表顯示，本研究最小的預(yù)測(cè)頻數(shù)是24.7，大于5，滿足假設(shè)5,具有足夠的樣本量。Chi-Square Tests表格也對(duì)該結(jié)果做出提示，如下標(biāo)注部分：Chi-Square TestsAsymptoticSignificanceValuedfC-sided)Pearson Chi-Square2.010Likelihood Ratiogms209Li nea-t)y-Linear Association7.7Ga1.OSM of Valid Cases150a. 0 cells (01.0%) have expected count less than 5. The minimum exp

16、ecled court is 24.67.即在本研究中，沒(méi)有小于5的預(yù)測(cè)頻數(shù)，可以直接進(jìn)行卡方檢驗(yàn)(2 X C)。那么，如果存 在預(yù)測(cè)頻數(shù)小于5的情況，我們應(yīng)該怎么辦呢？ 一般來(lái)說(shuō)，如果預(yù)測(cè)頻數(shù)小于5，就需要進(jìn)行Fisher精確檢驗(yàn)(2 X C)，我們將在后面推送的內(nèi)容中向大家詳細(xì)介紹。4.3卡方檢驗(yàn)(2 X C)的SPSS操作彈出下圖:嗆 Crosstabs-Layer 1 of 1n Display dustered bar charts Suppress ta&lesRow(S')：Exact. IuJStatistics. ICells. IFormat.Sty!殳”f

17、):J -1OK Paste 11 eset CancelHelp Kappa Risk© Crosstabs: Statist!cs-NominalOrdinaln Conti nig,ency coelTicient gammaPhi and Cramer's Vi . Somersd_ _ Lambda Kendall's lau-bC Uncertaintycoeifiaent Kendalls tau-c Norr'iinal by Interval! CorrelationsOChrsquare'-ni IT-nili m iii o o

18、f i匚Eta-| McNemar口 Cochran's and Maitel-Haenssel statisticsContinueJ Cancel j I H&lp© Crosstabs: Statist!cs3 Chi-squareI r “， ni'! CorrelationsNominalrOrdinal Contingency coefficient L i Phi and Cramer's V Lambda Uncertainty co efncientn Garnrnar _ Somers®dKendallf tsu-bKen

19、dall's tau<Ccntiniig I Cancel J Hgl衛(wèi) Nomimal 制 Interval .Eta Kappa Riskn McMemarJ Cochran's and Mantel-Haensiel statisticsIr ;t ?-,i匚osHdb畐 Cell DitSplay£ortinu&CancelHelp-C ountsjiiiB'r m mniii m rniii r HjOOserveC j mBiii “ niii Hl mi rHiiiBmaiif ptpected Hide small countsL

20、ess than 匸z4estt Coniare column proportionsig 匚-三 ic iirn Zill ell 匚 Unstandardized Standard'iized Adjusted stardardized-P erce nlages廠1 Row Columr TotalrResiduals-Noninteger Weights® Rourd cell counts Q Round case weightsO Tnjnc3l5' c&ll counts O Trunea1e case we(ghts © No adj

21、ustnentsContiI me CancelHeIp(5)點(diǎn)擊Percentage欄中的Column選項(xiàng)桔 Crosstabs： Cell DisplayCountsjiB - mI inBiii iiir rn maiir i-ij MiQbserv j 口 Expected二I Hide small counts二" II fl-Percentages "1 Row l3 Column TotalMest Compare coijrnn proportions.T匚=t > I 濟(jì)也H I E II III hl KrResidualsI j Un slaid

22、ardized Standardized Adjusted standardiz&d-Noninteger Weights Round cell counls O Round case weigh Is © Truncate cell counts ©Truncate case weights O' No aljustTients4.4組間比較彈出下圖:嗆 CrosstabsContinueCancelHElpRowCs): Display dust&red bar charts Suppress ta&lGEOK菱 riskjev&

23、lColumnCs;g intervention1 Lsyer 1 of 1Paste jl' KgsgtJ 、Cancel HelpExact. IStatistics.刑乩“ IFormat. |SMe- IC ounts 11 11 ! m I iirrniii n-n iiirTiii 11 iii ibii y conservedj H H . . . H X peeled Hide small countsLess than 匸-Z'test 點(diǎn)擊Cells，彈出下圖:n Compare column proportionsH I.JL ：= 3ll 三 II-H

24、：l I 1 Unsta ndardie 匸I Standard'ilzed Adjusted standardizedP &rce nlages gciw Column TolalrResidualsNonintsger Wsi gKit fi® Round cell counts O Round case weightsO Truncate c&H 匚ounts O Truncale ca&e weightsO No adjustmentsCounlsBI QOserved Expected Hide small counts11 匚-Percen

25、tagesO RowH Columri lotalContiU日 ncslHelp(3) 點(diǎn)擊 z-test 欄中的 Comparecolumn proportions禾R Adjust p-values (Bonferronimethod)選項(xiàng)鮭 Crosstabs: Cell Cispidy -z=te£t翩,Com£ar& column propijrtionsHAdjusip-valjes CBonferroni nnetiod|I Il llIlli“ IIWIIIBHBIIIHIIBIHII Illi-Residuals Un stand ar tod

26、Standardized Adjusted standardised-Non integer Weights Roura cell counts O Round case weights O Trtincale cell counts O Truncate casa weights G No adjustiTients5、結(jié)果解釋5.1統(tǒng)計(jì)描述在進(jìn)行卡方檢驗(yàn)(2 X C)的結(jié)果分析之前，我們需要先對(duì)研究數(shù)據(jù)有個(gè)基本的了解。SPSS輸出結(jié)果如下：risk level * iiitrwiTtion CrosstabiilationirtefverrtionTotalDrugDietExercise

27、riskjevel HighCountEKpectsd Count% within inlervention1雖26b30b7424.732.03624756.0%74.760.0%74045.3%Norrmal CountEjcpectiid Count% within inlervention34a22b20b7625.3£3皿253440曲25.34D.O76.050I代TotalCountExpected C(juni鳴 within irrbswention5050.0100.09t£050.0100.0%SO50.0100.C%160150.0100.0%Eac

28、h subsenpt letter denotes a subset of intervention categories whose colurnr proportions do not differ significantly lirom eacti other al the .05 level.該表提示，本研究共有150位受試者，根據(jù)干預(yù)方式均分為 3組。在試驗(yàn)結(jié)束時(shí)，藥物 干預(yù)組的50位受試者中有16位膽固醇濃度高，飲食干預(yù)組的50位受試者中有28位膽固醇 濃度高，而運(yùn)動(dòng)干預(yù)組的 50位受試者中有30位膽固醇濃度高，如下標(biāo)注部分：rishjewl * iiiterwiition Cro

29、sstabulationinterventionToialDrugDietExerciseriskj evelH<ghCount伽23b30b j74Expected Count24724.724.774J0I% wHhin intervention32.0%56.0%eoiio%49.3%McirmalCount34a22b如76Expected Count25 325325J75.0騙 wrthin inferventionee 0%MO%4QLD9tECU船TotalCount505050150Ejflpectd Count50.050C50.0IlSO.iO% within in

30、lerV'ention100.0%100lO%100.0%Each subscript IsHeir denotes a subset of intervention cate gories whose column proportions do not differ significantly from each other at the .05 level.由此可見(jiàn)，藥物干預(yù)比飲食或運(yùn)動(dòng)干預(yù)的療效更好。 同時(shí)，該表也提示，藥物干預(yù)組的 50位受試者中有34位膽固醇濃度下降， 飲食干預(yù)組的50位受試者中有22位膽固醇濃度下 降，而運(yùn)動(dòng)干預(yù)組的 50位受試者中只有20位膽固醇濃度下降，

31、如下標(biāo)注部分：risk level * iiTtervention CrosstabulationinterventionT<j1alDrugDietExerciserisfcjevelHighCourrt1頭2Gh3Db74Expected Count24.724.724774 0% wittiin intervention32.0%56g60.0%49.3%NormalCount34a22b20b76Expected Count：5.325,3ze.o% witliln Intenention69.0%44g40.0%50l7%TotalCounrt505050150Expected

32、 CountEO'.OEO.OGOO150.0% wittiin inteoention100.0%100 0100.0100.0%匚act! subscript letter denotes a subset of Inierv&rtion catsgarles whose column proportions do rot differ signilicantlyfrom each othsrat the .35 level.因此，我們推薦使用頻率但是，當(dāng)各組樣本量不同時(shí)，頻數(shù)會(huì)誤導(dǎo)人們對(duì)數(shù)據(jù)的理解。 來(lái)分析結(jié)果，如下標(biāo)注部分：riskjewel * irvtervetrt

33、ion CrosstabulatiorbinterventionTotalDrugDietExerciseriskjevftlHigliCounle38b30b74Ek”席祠 CQunt24.724724.774.0% within Inteiveniion320%E6.0%60 0%3.3%NormalCountW22b20breE)cpecte<i Count35/525.325376.0% within inteivenlionee.o%44.0%40 0%50.7%TotalCount刊50別150Ejcpected Count50.0501.0500150.0% within

34、intervention10010%1 00,0%laioo-b100.0%Each EubEcriplletter denotes a subset orinIsrvGrtion categories whose'column proportions do n 01 differ signlficanuyfrom each oilier at ire .05 level.該表提示，藥物干預(yù)組的 50位受試者中68%旦固醇濃度下降，飲食干預(yù)組的 50位受試 者中44%旦固醇濃度下降，而運(yùn)動(dòng)干預(yù)組的50位受試者中只有 40%旦固醇濃度下降，提示藥 物干預(yù)比飲食和運(yùn)動(dòng)干預(yù)更有效。但是這種直

35、接的數(shù)據(jù)比較可能受到抽樣誤差的影響，可信性不強(qiáng)，我們還需要進(jìn)行統(tǒng)計(jì)學(xué)檢驗(yàn)。5.2卡方檢驗(yàn)(2 X C)結(jié)果本研究中任一預(yù)測(cè)頻數(shù)均大于5,所以根據(jù)Chi-Square Tests表格分析各組的差別。SP SS俞出檢驗(yàn)結(jié)果如下：Chi-Square TestsValuedfAsynp, Sig, (2-sided)P earson Chi-Square9 1752.010Likelihood Ralio9 33e3Linear-by-Linear AssociationT.zea1.0C5N or Vai id Cases1503. 0 c&lls (0.0%.) have gstpeet

36、ed countless than 5, The minimum expected count is 24 67.卡方檢驗(yàn)(2 X C)結(jié)果顯示 X 2=9.175 , P = 0.010，說(shuō)明本研究中各組之間率的差值與 0的差異具有統(tǒng)計(jì)學(xué)意義，提示藥物干預(yù)與飲食、運(yùn)動(dòng)干預(yù)在降低受試者膽固醇濃度的作用 上存在不同。如果 P>0.05，那么就說(shuō)明各組之間率的差值與 0的差異沒(méi)有統(tǒng)計(jì)學(xué)意義，即 不認(rèn)為各組之間存在差異。5.3卡方檢驗(yàn)(2 X C)中的成對(duì)比較分析如果卡方檢驗(yàn)(2 X C)的P<0.05，說(shuō)明至少有兩組之間的差異存在統(tǒng)計(jì)學(xué)意義。SPSS俞出的risk_level * in

37、terventionCrosstabulation表格通過(guò)數(shù)字標(biāo)記提示了兩兩比較的結(jié)大家可能會(huì)注意到, 道組間標(biāo)記的作用即可。果，如下標(biāo)注部分：ri&h level * imetvBiition CrosstabulartioiiinterventionTotalDrugDialExerciserlBk_level HI 3.11CountExpected Count% within interventioniea2473223b24.7seo%30lb247CO.O%7474.049.3%Normal CountE?(peckd Count% within iniervention3

38、4aZ5.36G.09C,2225.344.0%30b25 340.0%76ye.Q50 了騙TolalCountE 坤敘 teM Count% within intervention50500lOO.OS*506D.01000%50501.01CO.O%1501SO.Q100.0%Each aubstrlpt denotes a subset af intrveniion cat&g心rkswPios豈 tolurrin proportions do not differ significantly fro rm eaiclh othsr at the .05 level.每組數(shù)據(jù)的

39、標(biāo)記相同 (即上下兩行的標(biāo)記相同)， 那么我們只要知fishjevel * iirterwirtion CrosstiabuldtioninterventionTotalDrugDietE3(ercisGfiskjevel HighCountExpected Count% within inteivention16a32 0420fc凋30b60lo47474.049.3%Npmal CcuntExpected Count% wHhin intervention34.25 368.0%22b25344 Q%20b2S.34010%7676.01507%TotalCountExpected Count矚 within inlerveition5050 0100 0%SO50010D.O%5050'0100.0%1501 50IJO100.0%Each subscriiF>t letter denotes a subset of intervention categories whose column proportions do not differ significantly from each other at the .05 Isvel.表格的標(biāo)記是什么意思呢？第另B么，risklevel * intervention