Medical Terminology Concerto
Method of statistic of clinical scientific research and Zhao Naiqing of choice Chen Shiyao
Writer unit: 200032 Shanghai, hospital of Fudan University Zhongshan is clinical epidemiology center
The processing of the data in clinical scientific research and statistical method are one of measure that clinical research concludes. And describe clinical data, using accurate statistical method is the premise that achieves result of accurate clinical scientific research. The article introduces the commonly used statistic method in paper of clinical scientific research, basic idea to the choice of statistical method from descriptive data.
One, the sort of data
1. data is phyletic: Does Qiu collect ǔ of a gathering of things or people of the that fasten neon?
Metric dataPoint to successive data, have particular numerical value normally, wait like height, weight, blood pressure, haemoglobin, bilirubin and albumin.
Grade dataPoint to the data that has certain level, be like:
Cent of clinical curative effect is cure, show effect, improve, invalid,
Clinical examine to be divided as a result for - , + , + + , + + + ,
The serious degree cent of the symptom such as ache does not have ache for 0 () , 1(is spent gently) , spend in 2 () , 3 (is spent again) etc,
Grade data calls half quantitative data again.
Group dataPoint to to there are order or grade relation between each data, be attributed to certain property however, can be two kinds, also can be many kinds.
Be like sexual data, press the male and female classification, computational male and female have how many; each
Professional data, wait for classify by worker, farmer, employee, the exemple with respective computation counts;
Along with seek result data, press live or dead classify, the exemple with respective computation counts;
Accept interpose measure data, by with experiment medicaments, with comfort agent classify;
Whether to exist smoking, drink, pylorus screw bacterium (Hp) infection expose an element, by smoking, do not smoke, drink, do not drink, classify of Hp (+ ) , Hp(- ) .
The category of 2. data changes:
If age data is metric data, but can press be more than 65 years old, 45 ～ 65 years old, be less than 45 years old of cent to be old, medium, youth 3 kinds and change is grade data or computation data. But those who need an attention is:
Metric data changeover is grade data or computation data is compared simple, but can no more change into metric data from grade data or computation data, accordingly, in clinical on collect data or computer to store when data, lv of take an examination is collected or store metric data, it is grade data or computation data according to needing to be changed again when data processing only.
Undertake to some index of two groups of patients statistical when examining, data is computation or grade data from metric changeover, statistical efficiency drops possibly also.
2, the statistical description of data
To clinical research data, normally we are impossible in the paper or the reader tells the specific feature of each patient in the report. We need to know this approves the feature of data, facilitate on one hand description, facilitate the difference of two groups of data or feature is compared on the other hand, this is statistical description.
1.Metric data (data) statistical description: Pool of ⒊ of Qian of rotten Rong of extensive of Nai of travel of 0 ㄖ takes ǔ of Wang of an ancient small-mouthed wine vessel of grandma of department of screen aperture persimmon
Central positionUse normally all count will describe, be like
Anhydride of the age of a group of patients, weight, haemoglobin, albumin, bilirubin, flesh and urea nitrogen,
The requirement is this class number occupies; of should obedient normal school
If normal school is shown after changeover of data classics logarithm, all can count with geometry represent its center position, be like
HBsAg drop is spent (18, 116, 132, 164) ;
To slanting voice data, represent its center position with median normally, be like
The limits such as the AL T when studying acute hepatitis, AST from tens of going to on 1000 fluctuant bigger, and the metabolic circumstance of every patient is abhorrent.
The data of normal schoolDisperse degreeUsable standard deviation will describe;
To slanting voice data, can use 4 minutes a limits (Inter2quartile Range, IQR) describes disperse rate, namely:
IQR is 100 minutes of the 25th digit (P25) ～ 100 minutes of the 75th digit (P75) .
2.Computation data (data) statistical description: Analyse persimmon shows Shen to cast Suo stupid tip attacks about instrument of of Cong of Mi of Jiao of Qiao bone Jiu seeks straight? of Ping of army Tao Qiu to callForm than, be like sexual data, in 100 patients, 60 males, 40 females, can express to occupy 60 % for the male, the female occupies 40 %;
If data cent is,do not happen, wait like death, sicken, calculable the index such as its mortality, incidence of a disease, show the strength that its produce, say to lead.
Form produce intensity than cannot explaining commonly.
Clinical on having a lot of rate is a scale only actually (form than) . If sicken rate is a scale, as a result of historical reason, still call sicken rate. So cannot will form come for incidence of a disease than mixing demonstrative happening intensity. When two rate undertake comparative, if some index enough is right the size of rate is influential (if the age is right mortality) , and the composition of this index (if the age is formed) differ in two inter block, need to undertake standardizing to this index.
The statistic of 3. grade data describes: Hook of miscellaneous mildew of tip of Song of Fu is killed like that Ke world alcohol attacks scrupulously and respectfully H Si Qiu brushs Pu Э to kill fact of of rice of ⒑ of province of of Wo of Xing regret to take pool of Fu of the 8 that employ hammer to treat ⒅ of curtain of Qu ⑶ truly small ⒅ ? is calculable of each degree. Divide with all sorts of degree patient number namely with total case load, each are formed should be than the sum 100% .
4. spends than number, odds ratio, opposite danger: Core of а of deceive ∫ Ting fears? is the index that exposes the connection intensity between element and disease.
Than severalBe positive rate and negative rate than. Opposite danger is spent (RR) is two kinds expose what the condition issues incidence of a disease to compare, if study the impact of Hp infection and cancer of the stomach, alignment of sex of the look up before using research, watch the occurence rate of a certain number of cancer of the stomach after year for 2 groups by Hp(+ ) and cent of Hp (- ) , the more opposite than estimating namely danger of the incidence of a disease that calculates two groups is spent.
Odds ratio(OR) is contrast in case of illness express to expose in research produce the relevant rate between with the disease, it is the approximation of RR, if study likewise, use case of illness to contrast design, it is two groups by cancer of the stomach and cent of blame cancer of the stomach, the Hp before investigating two groups of patients to come on affects a situation, the relation between evaluation Hp infection and cancer of the stomach needs to use odds ratio.
3, of quantitative data statistical examine
Two sample book should have in clinical research or of the difference between many example statistical examine, in order to decide difference is actual existence perhaps is caused as a result of error of sampling.
1.T examines with variance analysis: Miscellaneous Shen swollen mallet receives envy of establish mu if? of aperture persimmon is compared for two groups, usable T examines.
T examines have 2 kinds of methods, depend on the data is to become group compare or conjugate is compared.
The comparison of around treats to each patient in clinical scientific research: The around of level of the albumin after horn of of fellow of stew of ê of Shan of Qiang of ┪ of ㄒ of Xie τ beautiful holds Shan high to seek Ran? to use the remedy that protect liver is compared wait to belong to conjugate to compare more.
Clinical go up a lot of research cannot conjugate, be like new drug and contrast the comparison of medicaments, it is cure normally group and contrast group undertake group quite. When choosing T to examine, the method that these two kinds of T examine is different.
To two groups of above (much group) the data is compared, choose variance analysis of variance analysis; to also have 2 kinds of methods, depend on research design.
The influence that if study blood places time to determine to blood sugar,is worth, it is 4 to dividing each individual blood after exsanguinate of 8 healthy people, blood sugar chroma determines after placing 0, 45, 90, 135 Min respectively, this kind designs every 4 blood sugar to determine the value all comes from same and individual, call random section random, need divides poor analysis with random section.
Same, if our purpose is to compare 3 kinds to differ,fall the remedial effect of blood sugar medicaments, use randomization method to be patient cent 3 groups, every kinds of medicaments applies at a group of patients, 3 groups of patients‘ final blood sugar gives out to design variance analysis randomly completely quite. Those who need an attention is, content of each groups of example is equal with unequal when, as a result of method of calculation of sum of squares of across block deviation from average somewhat difference, what choose variance analysis computational formula is different also.
What need points out is, it is right that conjugate or design of battery of conpatibility of medicines compare group or the advantage that design completely randomly some element undertakes controlling, as above narrates result of 4 blood sugar, besides placement time, the others element same (come from same person) , reduced bias error thereby, improve statistical efficiency.
When undertaking variance analysis, when if be planted to A, B, C3,the curative effect of medicaments undertakes comparative, invalid hypothesis is 3 groups of curative effect identical, namely H0:A= B = C, difference has statistical sense and overthrow this when invalid hypothesis, the curative effect that its choose hypothesis H1 to be 3 kinds of medicaments fully is full not identical or completely not identical, can not distinguishing at this moment is curative effect of which two kinds of medicaments identical, which two kinds are different. A kind of natural think of a way is do not do variance analysis and do 3 T to examine directly, examine namely A = B, A = C, B = C, conclude directly from this. Look from statistical angle, this is incorrect, because it increased the first kind of mistake, namely the probability of false electropositive mistake. At this moment the α of significant critical value probability on statistic, already exceeded α =0. The standard of 05, cannot take consequently.
More reasonable method is make after variance analysis multiple quite (two two compare) . Multiple relative method is very much, commonly usedly SNK(Student2New2man2Keuls) law, Duncan law, LSD law and Dunnett law wait.
Applied T examines and the premise of variance analysis is:
When little sample book, no matter T examines or variance analysis, the requirement that logarithm occupies all should be normal school, problem of sex of data normal state can make an on-the-spot investigation with frequency graph or normal state examines;
Examine to ask with variance analysis the difference between each groups example variance does not have significance into group T () of variance neat sex. Neat sex of two groups of variance examines can examine with F neat sex of variance of; much group examines can examine with Baetlett.
2. data changes: After brighting flatter of Fu of Quan holds? of an ancient small-mouthed wine vessel of Xing Huan ㄌ of smile of Xi of of the larva of a tapeworm or the cercaria of a schistosome of thistle of naevus faint muching establish to pass certain changeover, can change into neat sex of normal state or variance, it is better to analyse the effect. Computation of radioactivity of time of number of the bacterium in be like water, unit accord with Pu Song to distributing, data can be not percentage of contagion sicken rate, leucocyte, lymphocyte to change rate, barium through; of square root changeover the rudimental rate that gastric platoon checks for nothing accord with 2 to distributing, data can turn over sine function to change; drop to spend a data to wait through square root can change through logarithm.
3. nonparametric statistic: Dispatch asks another name for Guangdong Province of Piao of cabin of official Bu persimmon of aperture of screen of Chinese torreya of Mi of rank hazel Fu is what the? that seek Ran examines when T or the premise condition of variance analysis cannot be satisfied and logarithm occupies is overall when distributinging to cannot decide or do not have proper changeover method, can use nonparametric statistic method. Examine at parameter accordingly, match pair of relative designs, nonparametric statistic uses symbolic grade to examine (Wilcoxon law) ; uses two sample two groups quite grade and examine (law of Wilcoxon Mann And Whitney Ranksum) or median examines design of battery of; conpatibility of medicines uses M to examine quite (Friedman law) ; much group uses H to examine quite (Kruskal And Wallis) . Express the choice that the 1 parameter when listing metric data is compared examines with nonparametric.
Express the parameter when 1 metric data is compared and choice of nonparametric statistic method
===========================================================================================
Devise a methodParameter statistic (note one)Nonparametric statistic (note 2)
-------------------------------------------------------------------------------------------
Conjugate is comparedConjugate T examinesAccord with examine, symbolic grade examines (Wilcoxon)
Two groups are comparedInto group compare T to examineTwo example grade and examine (law of Wilcox2on Mann And Whitney) , median examines
Group of conpatibility of medicines is comparedRandom section variance analysisM examines (Friedman law)
Much group is comparedDesign variance analysis completely randomly
H examines (law of Kruskal And Wallis)
==============================================================================================
Note one: Using a condition is data normal school, variance neat sex;
Note 2: Applied limits basically is slant voice data and data cannot be changed for normal state, grade data is compared in group.
Grade data also applies nonparametric to examine quite in group, h is used to examine when much group is compared, when two groups are compared, use grade and examine (Wilcoxon Mann AndWhitney examines) or median examines. If compare motivation of two kinds of stomaches to medication the curative effect with functional dyspeptic sex, curative effect evaluation is pressed show effect, effective, improve, invalid cent is 4 grade, two groups can be used quite grade and examine.
4, card just examines
The property that research group or a few groups of two data is qualitative or of classification, know commonly used rate or form than describing each groups feature. It is OK to across block is led or form the difference between the circumstance to whether have statistical sense quite just examine with card.
4 case of 1. express the card of the data to just examine: Boreal Ran narrow rank raft ㄐ says qualitative? of street of laborious of brandish of calamity of sell of neon of department of easy persimmon of shovel of persimmon of Gong analyse uses card to just examine normally, when if study Hp,infection and cancer of the stomach concern, group of case of illness of cancer of the stomach 100, hp affects 80 (infection scale 80. 0 %) , chronic gastritis contrasts group 100, hp affects 60 (infection scale 60 %) , whether prep above of rate of infection of Hp of group of case of illness of cancer of the stomach is chronic gastritis group, because error of sampling is caused,not be namely, statistical 4 case can be used to express card to just examine when examining. General computation Pearson gets stuck square, state two groups are formed only than differring, have general connection. In 4 case watch if data is less, theoretical value (the numerical value with each due check that gets according to disabling to be calculated suppose)<5, always watch number especially<40 when, or rational research number<1 when, need is used accurate (Fisher) inspection.
2.Travel list card just examines: Boreal Ran offers form of mallet ? or when the attribute of the data exceeds 2 to plant, this kind of form calls cavalcade the watch. Still calculate normally Pearson gets stuck square, express what contact commonly to examine, go namely variable and row variable are computation or qualitative data, concern without grade between variable each level, examining result expresses to whether connection exists between two variable only. If travel variable is nominal variable (qualitative) and when row variable is grade variable, can use nonparametric to examine, the trend gets stuck square or with notch on average all right difference undertakes examining. Pure Pearson card just examines to often cannot show an issue. To inter block in group the statistical sense that the card of the data just examines or interior makes different two rate difference examines, can use the card that increase advantageous position just examines or M2H card just examines. And if need storied element is very much the factor that perhaps affects a result is very much, and when statified and too much sample size cannot be satisfied again, returning to an analysis often is the option that research above all.
3. opposite puts those who lead a data to compare: ? of Fang of of Gan of of Ti of Meng of 2 take along sth to sb besides calculable year of survival rate is hand-in-hand outside getting stuck to just examine all right, still can be opposite directly live the curve undertakes comparative, use LogRank to examine normally.
5, the common error that method of the statistic in clinical scientific research uses
Apply accurate statistical method to be able to increase the reliability that research a result, and wrong statistical method often brings about incorrect research conclusion. The common statistical method error in clinical scientific research includes:
1. constitutes compared misuse: Show off of take along sth to sb all the data that clinical place of? of Piao of obstruct of wood of crane persimmon department obtains can be calculated only commonly compare and forming is not incidence of a disease. Form the intensity that happens than cannot showing an object normally, and form compared size to get the influence of a lot of other factors, the size that because this is formed quite,compares or application are formed when comparing demonstrative problem, cannot abuse. Only fore-and-aft along with the data that seeks research talent to get incidence of a disease.
2. interior makes the impact of pair of statistic index: When the curative effect that core of Qiu feel ashamed fears? comparing two groups of remedy or the prognosis that show two groups of patients, often need to notice other factor is right the influence of the result. Mark is changed or undertake to may affecting the factor of the result statified it is the best method that solves this one problem, if influencing factor is very much, the likelihood needs much element analysis to balance the influence of all sorts of elements. And the effect that ignores other factor gives wrong result possibly.
3. slants the misapplication of description of statistic of voice ration data and law of check proved recipe: You of thirsty of is troubled by rotten Φ of travel Nai extensive to shelter lognormal distribution of? of Fang of defect of of cover form emperor uses geometry to all count descriptive) , but the data that at present a lot of research report still is used only all count a description. Because all be counted,delimited momently only with standard deviation the feature of normal school data, need to express to all count ± standard deviation only to normal school data, but all counting ± standard deviation is not to slant the feature of voice distributinging data, should use median normally (75 % of ～ of 100 minutes of 25% digit 100 minutes of digit) the central position of score data and distributinging general situation. To slanting apparently
The across block of voice data is compared, t examines or variance analysis also is incorrect, should choose nonparametric to examine.
4.Conjugate (conpatibility of medicines) compare and become group quite: Wei Yu examines with two groups of relative T choose should consider to design according to differring, completely random design and design of battery of conpatibility of medicines also want to study according to differring the design is chosen, conjugate research design and the material that group of conpatibility of medicines designs belong to blame independence data, can use corresponding pair only T examines or group of conpatibility of medicines variance analysis, the material that become group of designs or designs completely randomly cannot (also cannot) with conjugate T examines or variance analysis method has group of conpatibility of medicines examine.
5. is package relative mistake: Ran of Chinese catalpa of goblet of Ran of hazel of Mi of miscellaneous to joke falling into oblivion disrelishs Φ Bi Rong to show tell Fu to treat Wo Pi dispatch to ask I are tasted treat? next reoccupy is corresponding multiple quite, and should not do directly all two two relative T examine or nonparametric examines, otherwise the first kind of wrong meeting increases. This one mistake often still can see on clinical research and magazine.
The condition that 6. statistic method uses does not accord with: Official of children‘s hair sluggish tastes joyous ǖ of Mei Xing of τ of ㄓ of any of several hot spice plants to if T examines to ask with variance analysis data is normal state,suckle Qiong? (or approximate normal state) distributing and variance neat sex, data of a lot of research is shown slant apparently voice still uses T to examine or variance analysis is incorrect. To be not the data that lose a value, if standard deviation is far big Yu Jun is counted, such data often is to slant voice distributings. Variance is neat the gender is very big to statistical result influence, want special attention. Be like the methodological choice that returns to an analysis again, because variable is what property,have to be in charge of and use recursive method in disorder, because variable is quantitative data to be able to use linear regression (or) applies after data classics is changed, because variable is minute of class number to be returned to according to can using Logistic, and live time can be returned to with Cox because of variable. The recursive analysis method that uses impropriety in disorder can reach indecipherable result.
The result that what concern with statistic should make clear in 7. paper: I attack screen of official of ⑼ of Jiu Jun of difficult of department of Ran of ⒈ of ā of any of several hot spice plants of Mei Mu coughs value of of idle of Miao ǚ street.
What final need points out is, study the accuracy of the result and research design are concerned, the choice of statistical method also is concerned with the means that collects a data, accordingly, statistical method is studying design phase makes right choice, is not when the consideration comes again after data has been collected. Otherwise, the reliability that research a result is suspected, and pure the alternative bias that depends on statistical method to do not have a consideration to considering to design and measuring sexual bias is cannot remedial.
Medical research medium statistical method mistake is used
Statistical content is very rich, the method of medical statistic is very much, every kinds of method has his applicable condition, every kinds of method applies to disparate test to devise a type each. The utilization rate of method of statistic of paper of our country medicine, show ascendant trend after 1985, but the count that the paper that medical magazine publishs puts in different level is wrong, the applied mistake of statistical method can make whole and accurate the research to give a mistake verdict that have. To reduce this one phenomenon, raise paper level, the problem that learns what respect of method of the statistic in scientific research paper often appears to cure below is Baconian arrange, list as follows:
One, did not use necessary statistic analyses a method or describe with statistic only
A few articles did not undertake necessary statistic is analysed, perhaps be opposite only study a result all number, rate undertakes comparative from size of sample.
2, the specific name that uses statistical method without the place that write Qing Dynasty or write far from
It is clear to answer to tell institute tell institute with statistical method in the paper, if explain not clear or grant to explain far from, the person that go over a manuscript or draft or reader are right of paper conclusion will not judge correctly. Conjugate design and the statistical method that become group of design data are different, if said to use T to examine only, judge its validity very hard; One is carried only in some articles " via statistical processing " hind, draw up conclusion; Some is flat even do not carry " statistic " 2 words, explain with P value the problem finishs sth directly.
3, the data slants badly voice uses T to examine however or variance analysis
T examines and data of variance analysis requirement is subject to normal school, and variance is neat, the much data in medical research is disobedient normal school. Should distributing when deviate normal school is not old, not big to the influence of its result. But still ought to do normal state sex to examine first to metric data, if normal state sex examines to identify amount to to occupy disobedient normal school as a result, can have variable shift, or have nonparametric statistic. Can see from the data in the paper its data is serious sometimes deviate normal school, but still use T to examine or variance analysis. Because medicine research data cannot be negative number, when example not quite hour, average decreases 3 standard deviation should not be negative number, otherwise with respect to deviate normal school rule.
4, T examines the comparison that replaces variance analysis to have much across block
This kind of phenomenon is not scarce still, be in statistical when the comparison of on many groups of metric data, ought to do first examine always (variance of each across block uses variance analysis together, variance needs not together to handle) with nonparametric statistic method, on the foundation that reachs the difference has statistical sense, redo is multiple quite, if every two groups all are counted,become relative Student-Newmn-Keuls law, or many processing group and same contrast group of relative Dunnett laws, and the some in comparing K group is opposite or some a few pairs have all figures LSD way of special significance to wait on major. Common error is in the article ravel the data, fall to all sorts of combination two two all count respectively make it group designs two sample book relative T examines or conjugate T examines, and every time relative examine level still is =0. 05, can increase the probability that make a mistake so, it is the difference miscarriage of justice that does not have statistical significance originally have statistical sense.
5, examine into group T replace conjugate T to examine
Randomization is the important step that makes sure blame processing element is balanced and consistent in group, raise a test group and contrast of across block can compare a gender. The purpose that conjugate designs also is to reduce jumbly factor the influence to tackling an element, it tackles an element than becoming group of designs to be not more balanced and consistent, both key is the experiment designs plan to differ, analytic goal is different, its statistic method is different also.
6, the variance analysis that section designs is replaced repeat the variance analysis that measures a design
Repeat measure a design to look be like random section design, but what differ with its is test result arrange by time order, the processing of random section design passes unlike to be arranged randomly in that way, its are relevant between different time, not independent, not only can analyse two elements whether is there a difference between each level, still can analyse two elements to have without interact.
7, one-way and orderly variable is done examine
Clinical be duped curative effect or examine to be divided into many grade as a result, if curative effect cent is heal, show effect, progress, invalid 4 grade, criterion Person can examine only each groups are formed identical, and cannot examine whether do each groups of curative effect have a difference.
8, misapplication examines formulary
The formula in examining is more, each have its applicable condition, have a bit inadvertent, have the possibility of misapplication namely, answer to undertake choosing correctly according to the property of experimental design and data. Common error has:
1. is common 4 case express a data, when N>4O, but when having 1<T<5, corrective without computation Y is worth.
2. is common 4 case express a data, when N<4O, or when having T<1, still use examine, did not choose 4 case to express exact probability law.
3.R×C expresses a data, rational research the case of several T<1, or the 1 ／ that the division number of 1<T<5 exceeds total squares formed by crossed lines to count 5, did not use proper processing technique, and R of direct apply mechanically×C expresses examining formula, bring about an analysis slant quality.
Data of 4 case watch arranges 4. conjugate for common 4 case watch, both designs plan to differ, the meaning of A, B, C, D is different, analytic purpose and method are different also.
9, linear and relevant the problem in be being analysed with linear regression
Undertake linear and relevant when be being analysed with regression, reach recursive equation or cipher out R value, before concluding, answer to do hypothesis to examine first, in order to concludes the relation of depend on sb or sth for existence that whether linearity exists between variable or correlativity, see the size of R absolute value even as to relevant close degree, because the hypothesis of R examines, no matter how P value is little, can show variable only relevant, and the information that cannot provide relevant and close rate. R absolute value is close to the correlativity between variable more closer. Call decision coefficient, express to return to sum of squares to hold the proportion of total sum of squares, there should be correlativity between variable, but not be very big when, the correlativity between clew variable is actual the sense is not great, worker of some scientific research lacks understanding to this, ever discovered in the paper R value is 0. 126, p<0. O1, decide coefficient =1. 59, and did not cause investigator the care of real to its significance. Return some using linear and relevant replace a curve relevant, with linear and relevant replace grade relevant, quantity of meet an emergency uses linear regression however for 2 classification variable.
10, the problem in much element analysis
As the computer gain ground, much element analysis already applied at medicine widely increasingly in research. The much element analysis that uses in medical research has model of risk of scale of multivariate and linear regression, Logistic regression, Cox, differentiate analysis, clustering, advocate analysis of regression of law of composition and factor analysis, typical relevant analysis, corresponding analysis, Multidimensional Scaling, Poisson. As a result of the complexity of these analysises, some investigator are very not familiar to analysing medium criterion, lack the basic knowledge of statistical principle, to choosing what data, application what computation and how to explain earnings result to wait, sheet relies on the computer to be finished cannot satisfactorily entirely. Because lack statistical basic knowledge, machinery uses statistical software, bringing about the result that obtaining the computer to give out not to know is what meaning. When undertaking statistic is calculated, often need to use statistical software neatly, this has more thorough knowledge with respect to the computational method that needs pair of software. People is handling " data of statistic of much element much index " respect most the mistake that often makes is:
1. is multivariate (or much element) the data uses a yuan (or only factor) processing of statistical analysis method: The utilization rate that can cause a data so is low, cannot reflect the integral condition of the data, cannot announce well the interact between variable and immanent connection, reach easily one-sided, even the conclusion of subreption.
2. much element analyses choice mistake of the method: In analysing paper of our country medicine to much element, use most is multivariate regression, commonly used multivariate recursive method has multivariate and linear regression, Logistic regression and Cox regression, the type that measures according to meet an emergency will classify them, quantity of its meet an emergency is respectively successive model variable, classification variable and live time. If there is many observations index in the data, but the fraction that measures without independent variable and meet an emergency between them, when studying the far and near between variable concerns, can choose variable clustering; According to the relation between variable, want suffer when trying an object to undertake classified, can choose sample clustering; Want to reduce variable dimension, when with a few a few integrated variable conveys the majority message that numerous former variable reflects, need to choose advocate composition analysis or factor analysis; Mirror variable and sample when a rectangular coordinates is fastened at the same time, should choose corresponding analysis. There should be classified variable in the data, still have a series of mensurable observation index, if want to classify variable and different level quite only of the mensurable index that many have certain connection on major when all counting the difference between to whether have statistical sense, can use multivariate analysis of variance; If classification is variable the representing‘s is a few clear classification is overall, the hope establishs a kind of method, right sealed and individual when having attributive judgement, should choose differentiate analysis; When if be in,doing multivariate analysis of variance, discovery still has the influencing factor of one or more ration, the hope affects his multivariate analysis of variance of the redo after deducting, right now, should choose multivariate covariance analysis.
3. independent variable has in all linear mixes with multivariate and linear regression multivariate Logistic regression: In all linear is to point to to there is dependency between independent variable, serious in all linear meeting makes return to equation to be not stabilized, if make the action of independent variable and actual and contrary, the independent variable that has statistical sense turns into to wait without statistical meaning. To multivariate (or much element) data, a lot of articles analyse a method to choose independent variable with only factor, built again next multivariate recursive equation. Actually, much element analyses itself to be able to choose independent variable, using only factor analysis to choose independent variable is wrong, use only factor analysis to cannot be taken more only. In fact, what people often faces is the complex statistic data of much element, nonexistent some is planted statistical method can use full data, answer the whole question that expects to solve on major. This combines major and statistical knowledge with respect to need, choose different variable subclass, undertake all sorts of corresponding statistic are analysed, make professional knowledge and statistical knowledge marriage, make reasonable analysis and explanation to the result.
The place on put together is narrated, medicine puts the statistical error that be in in scientific research paper, not be abstruse mathematical problem for the most part, it is statistical ABC even partly quite. Want us to strengthen medicine only statistical study, lay next solid statistical foundations, can decrease statistical the error that go up, the article that makes those are put in statistical problem disappears from inside the medical periodical of our country, paper compose more accord with scientific sex and rigorous sex requirement.
No comments:
Post a Comment