Professor Daryl J. BEM, a professor of social psychology at Cornell University, believes，People can "predict" what will happen in the future。2011，Bem tests whether it exists through nine standard psychological experiments。In the nine experiments of this study, there are eight results of "support" the existence of this predictive future phenomenon。So，How did Bam infer this conclusion? Like most social sciences and psychologists，He uses a null hypothesis significance test, nHST)。For example，In the first experiment，100 participants must first use the button to guess which position on the computer screen (left or right) will appear.。To test his assumptions，Bem sets the zero assumptions (H0) in the NHST as "the probability of the participant's guessing position is equal to 50%，is determined by the level of random probability "; and the hypothesis (H1) is the probability of guessing 50%。Data analysis，They get p = 0.01，threshold 0.05 less than NHST，Refusing zero fake，So as to get selected assumptions，It is confirmed to predict existence。

Bam's research results have caused a huge response in the scientific community。The reason is that predicting is a very abnormal phenomenon，Human beings have not found evidence support in biology or physics。But the method used by Bem is the "standard" statistical inference method in social science and psychology。So，Where is the question?

Researchers have questioned more about the NHST method。Actually，Researchers who have criticized this method for a long time，However, bet365 Play online games their voice has been ignored for a long time。After the papers of Bem，There is a way researchers have clearly pointed out，Psychologists need to re -consider whether to use the "standard" NHST，Because this kind of statistical method is too easy to be affected by other factors，It is easy to let researchers deceive themselves，also deceived readers。

Although NHST is the most commonly used statistical inference method currently used in social science research，Researchers usually want to get the results of P & LT; 0.05 to prove the research theory，But this may lead to the problem of the publication bias。That is to say，When the research results show that P & LT; 0.05，Thesis is usually published; on the contrary，p≥0.05 The paper that cannot refuse to refuse the fake is usually not published。This is here，Readers may only see those studies that get significant results，As mentioned earlier, a large number of unpublished research conclusions have not been displayed，For example, other researchers repeatedly repeated Bem's research data that did not obtain significant results。This screening mechanism will mislead readers' awareness of research problems。Psychological repetitive research project shows the consequences of the publishing deviation。Only about 36 % of the repeated experiments of the 100 psychological research that have been published have confirmed the results of the original research。The core of the issue of the deviation is the pre -pre -pre -pre -pre -pre -level。

Under the crisis of repeated crisis，Researchers recommend considering the alternative of NHST: Bayesian factor test。The following will be briefly introduced in social science research，How to use the Bayes factor to help solve repeatable problems。

Bayes factor is the Bayesian bet365 live casino games hypothetical inspection indicator，It also needs to define zero fake and selected assumptions。In the predicted research above，The probability of the zero fake is 50%，that is, H0: predicting does not exist;。Bayes factor BF0U quantitative data that is more likely to be observed by assuming H0 by assuming H0，In other words，BF0U measures the degree of H0 relative to HU's data support。For example，BF0U = 5 means that the support of data for H0 is 5 times the HU，The evidence of "no prediction" is "predictive" 5 times the prediction，The evidence of "predictive" is "no prediction" 1/5。

One question that researchers often ask are how big or more hours of BF0U，It will accept or refuse zero fake。For example，How big or small Bayes factor is obtained based on experimental data analysis，Can confirm the predictive existence。Behind this problem is the demand for deep -rooted threshold root，If α = 0.05 in NHST can decide whether to refuse zero fake。However, unlike NHST，Bayes factor is a continuous value，Generally do not make two points (rejected or not refused) judgment，Instead, quantitative assumptions are supported by data support。If BF0U is near 1，There is no preference for zero fake or selected assumptions，That is, the Bayes factor cannot be determined，It is likely to need more data to prove which assumption is correct。Another more direct question is，When the Bayes factor should (or small)，Journal will be published by the article? As early as 1961，Harold Jeffreys pointed out，BF0U & GT; 3.2 or BF0U & LT; 1/3.2 indicates that the data has positive evidence to support H0 or Hu， BF0U & GT; 10 means strong evidence。1995，Kass (Kass) and Rafeitery are recommended to use BF0U & GT; 3 or BF0U & LT; 1/3 to indicate data support Bet365 lotto review H0 or Hu。But what you need to pay attention to is，These thresholds also have no strict theoretical basis，Using threshold can also cause problems with the deviation。In fact，The setting of the threshold is artificial，or even subjective。We recommend not setting thresholds，Instead, according to the actual size of the Bayes factor，Researchers can make the degree of inference conclusions that H0 is supported by data.。Do not make a two -point judgment and abandon the use threshold，It can make the Bayesian factor test avoid the irreplaceable problems of social scientific research to a certain extent。

NHST departs from zero leave，Discuss the possibility of its rejection，The result can only be refused or not refused zero fake。What needs attention is，Zero -fake setting does not reject it does not mean that zero vacation is accepted。So，Regardless of the results of the NHST test，None of the researchers can get the conclusion that it is not predicted。Researchers either get the conclusion of the research theory of refusing zero fake proof，Either you can't get any conclusions (research cannot be published naturally)，This is also one of the reasons that lead to the deviation and repetitive issues of publication。When using the Bayes factor to evaluate the zero fake and selected assumptions，The status of the two assumptions is equal，It does not require the traditional "zero" assumes that the zero fake is true。Under the framework of the Bayes factor，H0 and H1 are just the assumptions that two researchers care about。Observation data combined with prior information，Beyes factor can get two relative evidence supported by data。This means that the Bayes factor can determine that the "selection" assumption Bet365 app download is better than the "zero" hypothesis，It is also possible to get the conclusion of the "zero" assumptions than the "selection" assumption。Place the zero fake and the preparation assumptions in the same position，When the data supports zero fake, it can also get effective conclusions，Articles that make the analysis results "not significant" may also be published，It also helps to study repeatability。

Bayes factor measured the evidence of data support or opposition to a certain assumption，and the P value in NHST does not have a direct interpretation of evidence (P & LT; 0.05 just means that the possibility of the current data or more extreme data is very low if the zero fake is true)。From this perspective，P value will exaggerate the evidence that refuses zero fake。For example，Nine experiments on predicted research，Researchers use NHST and Bayes factor analysis data。NHST results display，Nine experiments have eight results from P & LT; 0.05 results，Refusing zero fake，It can be inferred that it is "confirmed" that the predictive existence exists。and the Bayes factor only has 3 bias for selection assumptions (bf0u & lt; 1)，and none of the BF0U & LT; 1/3，There is no enough evidence to indicate that the zero vacation should be rejected。So，Bayes factor test for nine experimental data，No evidence is found to indicate that predicting existence，This is completely different from the P value -based NHST conclusion。The reason for this result is that the P value is too easy to refuse zero fake，Conclusions that produce unreasonable conclusions。

Many software can achieve calculation of the Bayesian factor，Including the BayesFactor，Bain， bfpack，Bayeststr), etc.。These software can automatically set Bet365 lotto review parameters，Researchers only need to enter data and specify the model used by data analysis (such as T test、Formula analysis model)，Research hypothetical care (such as H0: μ1 = μ2 = μ3，Among them, μ represents the mean value of each group)，You can get the Bayes factor that the research assumptions。It is worth noting that，BayesFactor and Bain software package can be used in the Jasp visual statistical analysis software to complete the data analysis through the mouse clicks and drop -down menu，It provides convenience for the Bayesian factor test for the social science and psychology researcher who is not familiar with R language programming。

Although we do not recommend using the Bayesian factor threshold for assumption，But researchers may want to get a clearer conclusion during a data analysis。At the same time，Although we believe that the use of the Bayesian factor test assumption that can avoid publishing deviations and uniquely studying problems to a certain extent，But different parameter settings、Software selection will still lead to different analysis results。To overcome these problems further，PREREGISTRATION is a formal recommendation of the current researcher。Ideally，Pre -registration requires Researchers to write an article before collecting data (that is, the pre -registration report)，Explanation of research theory、Experimental design、Data Analysis Plan，But there is no need to present data description、Data analysis and conclusion part。Based on pre -annotation report，Journal will determine whether the study has published value。If the pre -registration report is accepted，Researchers will actively carry out data collection and analysis work，Report the conclusion，bet365 live casino games No matter how large the Bayes factor gets，Articles will be published。Current，CENTRE For Open Science has provided pre -registered registered registered and many pre -registration templates，There are also many important journals encourage pre -register research，Such as psychological science、"Journal of Psychology", etc.。

　　Bet365 lotto review