 Let me welcome you to my third lecture here. Now, in this third session I would like to draw your attention to something like psychology as a science. Usually in psychology we teach students to conduct experiments and look for significant findings. By significant findings we mean that your test must be significant at point 5 level of significance. Now in all social sciences this is a major obsession that can we measure things, can we quantify things or not. And in that process I have learned something that there can be circumstances where you do not get a significant effect. Nevertheless those non-significant effects are meaningful and interpretable. So, I would share some examples of such situations with you. So, that we develop respect for null findings. And in that context first I take you to the work of other social scientists here and here are two quotations for you. Let us look at the first one. .. So, the first quotation you see here what is the attitude of people toward measurement. They say when you can measure what you are speaking about and you can express it in numbers you know something about it. But when you cannot express it you cannot measure it and when you cannot express it in numbers then your knowledge is of a meagre and unsatisfactory kind. That means we have to measure things and quantify things. This quotation is from William Thompson. But if we go to a famous economist who got Nobel prize he also had a very similar advice. And his advice was that if it matters measure it and if it cannot be measured measure it anyway. So, measurement is a sign quenan any social science which is trying to be science has to use quantification has to use number here. And that is why we train our students to conduct a study demonstrate that the effects are statistically significant. But in this process let us go back to something about human mind. And three different kinds of views I would like to share with you. One view is like John Tindall. Let us look at his background. He is not a psychologist. He is a physicist, a mathematician and a chemistry man. He is the one who was offered the chair of natural philosophy at the Royal Institution of London. And he succeeded Michael Faraday as its director in 1867. And he was one of the founders of the journal Nature which is considered to be of high prestige value. What did he say about human mind or measurement or causation. And I thought this is the most persuasive quotation I can find for you. Every occurrence in nature is preceded by other occurrences which are its causes and succeeded by others which are its effects. The human mind is not satisfied with observing and studying any natural occurrence alone. But takes pleasure in connecting every natural fact with what has gone before it and with what it will come after it. Now that he wrote in a book in 1872 which is titled Forms of Water, Clouds and Rivers, Eyes and Glaciers. 12th edition I have taken this quotation. As a psychologist I find this is a profound statement about human mind and we ought to be following his advice. Now on the other hand we have another view which came from behaviorism in which we said if we are studying cause and effect relationship all we should be studying it what we can observe. So what we manipulate are observable what we measure are observable why should we go in between the two. So anything which is unobservable do not allow us to become a scientist. So a skinner called such process a black box here and this view lead to what we call a stimulus response view of what you manipulate observable what you measure observable here and relate these two. And that lead to a wider acceptance of analysis of variance in psychological research. And few things here we have started testing any causal hypothesis. Tindall said we always make causal connection between what preceded and what would follow. So we have started using analysis of variance to see whether the cause produced any effect or not. And that effect we will test by a statistical significance of the either the main effect or interaction effect in analysis of variance. And to judge something as a statistically significant our cutting point was 0.05. So if something is significant at the level 0.05 we say it is truth anything which is more than 0.05 we say it is a null finding it is nonsignificant finding it is not to be given any credibility we usually ignore it. I am concentrating on such null findings what happens when your effects do not reach 0.05 level of significance. So do you discard them or you should interpret them this is the goal I have said for this session. And for this we have to take a third view which said that simply a stimulus and response are not enough we have to consider organism between these two. That lead to the notion that effect of a stimulus what is observable on the response which is another observable is actually transmitted by the organism which I will call here latent variables or implicit variable or mediators here. So three views we have tindal saying we have connections a skinner said studied the two observables Woodworth says that no between the two observable we should also infer what happens in between these two this is what is the goal now to account for this I am raising two issues here. Let us look at the national scene in this country the way we allocate funds in our budget and the way the fund would be reaching the beneficiary. So our former prime minister siri Rajiv Gandhi he made a one statement which I have produced here from internet Rajiv Gandhi once said that only 15 paise of a rupee that is the effect size is low that is allocated for the rural areas which is the independent variable or I V reaches the true beneficiary that is the dependent variable. Rajiv Gandhi now says that it is only 10 paise it is only 10 paise means the effect size is really low and this is what I am debating. And then here is Devendra Sharma who said we read how money is being squandered in the name of development how the traffic police makes money how the government officials move the files and how the public services have to be paid for and I am calling these as mediating variables. So budget allocation we have beneficiary we have and what goes in between are the mediating variables. This is what public policy researches are doing and psychologist can be of great help and that is why I selected this example. Now this is the challenge to a psychologist how can you represent this phenomenon which is prevalent in this country how can we explain this small effect or null effect to do that I would give you few examples of here. But let me do one exercise here what do you see here if you show this picture to different people and ask what do you see different people would come up with different answers for example if you would ask a group of philosophers like professor Mishra he would say logically you could not say that they are crossing the road as the duck does not know what is road and how to cross it. Now you ask another person who is a scientist he would say with this speed and width of the road they will cross the road within 15 seconds. But if you ask you see we have most of the our audience female if you would ask feminists they would say why must the mother duck bring the duckling to cross the road but not the father duck. How did this answer come we have manipulated picture we have asked for their response. So between the two observables people are coming with different explanations and if I am a psychologist if I am a behavioral scientist I should be able to tap this process latent variable their interpretations and we have methods through which it can be done now. So these questions are apparently coming from the implicit knowledge responses which we have to infer or this position of the individuals. Let us consider how we are going to do so two things new methods new analysis I would like to share with you before I give the examples. The first one we had two major analytic tools available to us to make psychology as a science one was analysis of variance or ANOVA another is called regression analysis just make a small comparison and contrast of these two. In ANOVA what do we do we manipulate the x we measure the y and the effect of x on y we represent by this c that is the total effect. Now those who believe in SOR view they became dissatisfied and the example I gave you that people come up with their own implicit disposition their own implicit knowledge. So they said we must divide a method which can tap that process. So here we need to consider three things x which is known variable y which is known variable but the implicit variable is m which is mediator. So this c you see is the total effect of x on y which is the observable effect now the question is how this c can be partitioned into c prime which is the direct effect of the i v on the d v and how can we estimate effects of effect of x on m and effect of m on y that we represent by path a and b that is a simple two step regression analysis in the first stage you predict y from x second stage you predict y from x and m together. You have to do another simple regression analysis in which you predict m from x. So once you have these you would be able to construct a model like this. So a times b would become the mediating effect the effect of x on y through m effect the c we already know we have been knowing through analysis of variance. So regression analysis gave a new tool a new perspective in tracing the implicit variable that what carries the effect of x on y to what extent the effect is of x by itself and y to what extent it is coming through the m. So with this technique Baron and Kenny when they proposed they said some conditions and let us look at those four conditions the first requirement was c which is the total effect must be greater than 0 that means it should not be non significant and I am dealing with even though it is non significant we can do it this is the departure I am making even though it is non significant this technique is useful according to Baron and Kenny if this is not true you need not do anything people would say what effective are trying to study to be mediated I am saying it is not necessary that x would have effect on y still you can interpret it and see how we can do this. So I would like to make three main points that if we consider the development in SOR and regression analysis technique which gave simultaneous equation modeling we would be able to deal with such phenomenon which we consider to be irrelevant non significant useless examples of poor research. So first point I would give you four examples in which the effects are not significant but they are meaningful and interpretable. Second thing I would say that with these with proper conceptualization method and analysis you can make those effects as meaningful as any statistically significant one from the stimulus response view and final point is if psychology is ever to be of use to the government to the society it can be able to give a good answer for most of the policy analysis. So the issue I raised that why 15 paise has become 10 percent now 10 paise now from the budget a location we should be able to explain using our method this is what I have set as my goal. Let us come to three four cases here in part three the first case it is a very interesting and humorous example you have seen let us suppose you get a consulting project from one of the call centers and they feel that if they hire intelligent employees who would be attending to the call from US UK Germany so on and so forth they would be doing good service and would be making less error. So the hypothesis is that intelligent people would make few errors while responding to the customer calls from the US that means if you calculate correlation between x and y that correlation should be significantly negative greater the intelligence less the error by the employees of the call centers. Now you conducted a study in Gurgaon with 200 people measured their intelligence and the errors made by them and got 0 correlation a degree of freedom of 198 what to do with it would you write in your consulting report intelligence has no relationship with error made why are you wasting your money over measurement of intelligence a stimulus response model would say the matter stops here SOR approach mediation analysis is no there is a problem this effect means something and what it means I show you next. Now another researcher instead of a bundling his research program he thought there was some problem why I conceptualized my research truth is intelligence leads to monotony monotony leads to increased errors. Hence negative relationship between x and y that is intelligence and error is suppressed by monotony. So if we measure three things now intelligence monotony and errors you would be able to find out what you wanted to demonstrate and precisely when we did it and what I am telling you is nothing profound this idea was given in 1979 and later on in 1991 psychological bulletin but somehow we psychologist lose sight of these things anything which does not fit within our model we ignore it so let us look at the same example now what I had shown before let us look at this data relationship between when you conducted this experiment relationship between IQ and error is still 0 0 like your previous study fine look at the chart now but this time IQ does lead to monotony this regression coefficient is 0.65 and when you predict error from IQ and these this value is 0.75 and this C minus E C prime is minus 0.49 is in this what you had predicted that relationship between intelligence and error would be significantly negative and you have it this effect was nullified because of the suppressing variable of monotony go to the chart and see and I have given here numerical example to see and in 2002 Sraut and Bolger said it and they actually challenged Baron and Kenny there is no need to have significant effect of x and y to do mediation analysis and here is one example I have given you got it now let us come to another example these days second case I am illustrating in India 80 percent of the research by the psychologist deals with a stress and workers are overloaded students are overloaded those who are working for what we call the other service the service when we send outside subcontract yes they are working 12 hours from 8 to 8 p.m. they would work a students bag you would say everyone is going like this. So, if you are asked to a study by ministry of human resource development what is the impact of workload on health of the students or employees how would you do the study now let us suppose you did this study and the Sraut and Bolger example I give you we have to consider both inhibitor and facilitator something augments something suppresses the first example I had given you of monotony watch suppressor this one I am giving you the two examples. So, now let us suppose here coming to this case now workload your hypothesis is workload has adverse effect on employees health again negative correlation negative beta slow. So, if you manipulate x and y then there should be main effect that is high workload to worse health then. So, greater the workload poor the employees health this is your hypothesis and when you manipulated workload among the workers and took measures of sound health how many times they report how many times they take leave on the basis of health reason so on and so forth. You came up with your finding you know. So, you took this measure after one month now in between you also took measures of how much stress they experienced and also how many times they went to temple did exercise took vitamins so on and so forth two things you measured. So, what kind of coping activities they engaged into one feeling of stress another is what kind of coping activities you have initiated to deal with it. So, once you have taken it and when you analyze the effect of x and y again your c was 0 effect of x on employees health was non significant not even it was 0 just like the first case relationship between intelligence and error is 0 effect of workload on worker and the student health is 0 would you say ministry increase load it is has no effect no the two other variable you have measured would help you answer the question look at this chart workload has no effect on health 0 c. Now, workload is increasing stress look at part a 0.76 and a stress is making them this has a negative effect on health higher the stress poor are the health poor measure we have. Now, look at the bottom workload increases coping device good food physical exercise regular swimming vitamins going to temple so on and so forth and coping leads to good health. So, take the product of the a and be in the first part that is a negative value of minus 3 0 and this one is a positive value coping has a positive effect on health a stress has a negative effect on health that is your prediction and when you enter these two please look at your my c prime now when you have control for the facilitator that is coping device and a stress which is inhibitor on negative your predicted effect negative relationship between workload and health is significant that is regression coefficient is minus 0.18. So, in a experimental study when there are two mediator one is facilitating another is inhibiting you may get total 0 effect no effect if both are equally powerful if inhibitor is more a stronger than facilitator then you would find some effect, but weaker or vice versa something like this. So, here is another way that even though your effect is non non significant or null they make psychological they are psychologically meaningful such findings we need to consider such implicit processes or latent variables here. Let us come to a third case these days people talk about moderating variables. Now, moderating variable means here I give you just one example leaders are supposed to be fair they are not supposed to be unjust. So, if you do one study in which you manipulate be a leader was given a chance to do some distributive justice distribute things between different subordinates. So, we believe he should be fair he should be objective and fair, but one leader is fair another is unfair whom would you punish more this is the experiment I did this is our own study here. So, our idea here that people would generally lie punish unjust leader more than a just leader. Now, I am bringing a moderator when would you follow this rule this you would follow without group if somebody is member from the other group, but if he is of your own group many examples in this country in political scene organizations you may have seen this rule is true when you are dealing with out group member with in group member you may not punish ignore it. So, we are predicting interaction between categorization and leaders behavior fine. So, leader punishment is the dependent variable categorization into leader behavior are our IVs we are predicting interaction. Now, in between we have measured a number of and it is very interesting if you find only effect main effect of leader behavior then you would say that people are fair because they do not go by categorization. If you find effect of it then we will say we have double standard in group we do not follow this group with out group. So, there should be interaction. So, now I am demonstrating it if you in experiment if you find no moderation means non significant interaction effect this non significant interaction does not mean that you are a poor researcher. What it means that some suppressing variables are taking place and here one example I give you what we did in this experiment we manipulated distributive injustice and in group versus out group managers by gender main and women at the same time punishment is at the bottom our dependent variable we measured outrage how angry you feel how embarrassed you feel how mad you feel with this leader how happy you feel with this leader because if in group does favor you would be happy out group you would be angry then we measured this positional attribution is this leader this kind of person objective type in group every type we also measured external attribution. So, when what is your attitude are you going to support or you are going to take a position. So, attitude as another mediator we measured now when you would look at these and look at the interaction effect on outrage we have significant interaction on external attribution we have significant interaction on attitude we have significant interaction, but on your d v the interaction is not significant point one zero. So, by traditional standard your research is gone you would not get your degree right this is the issue we have to meet this is not true this categorization or no moderation is an outcome of separation of by outrage and mediation by external attribution and attitude my leader in group has done because of external situation no rule regulations because of favorable attitude toward them this is what I have demonstrated here and let us look at the interaction effect here if we look at the plot the four interaction graphs your d v is here and your m v's are here you see the reversal pattern like just look at the solid line and open line width. So, here and here you see one kind of pattern here you see reverse kind of pattern outrage, but external attribution in attitude like we are protecting our in group, but in anger like if my in group has done something wrong I feel very angry never the less I protect this kind of two kind of forces are operating in my decision making and when we do this now let us suppose when you would look at this I am simply saying just look at the slope of the two lines anger is a stronger in case of in group, but external attribution and out group is weaker you see the slope cellular slope in case of the in group when it comes to punishment it is with the out group more. So, because of these contradictions your interaction effect is not reaching point zero five, but it is possible to bring it back to point zero five and how do we do it if we can consider I have done three simultaneous equation modeling here in which the predictor is here the first one you say the interaction between behavior and categorization and two main effects I am controlling this is the requirement whenever you have interaction. See the first one original effect one point two zero interaction effect which is not significant once you control for outrage that interaction is becoming now significant two and that was your prediction the interaction should be in punishment when you come to external attribution see one point two zero is being further reduced. So, external attribution is a mediator when we come to attitude the same thing you are noticing. So, attitude and external attribution was doing moderated mediation in case of categorization and leader behavior, but outrage was doing outrage was doing suppressing role. So, we feel angry with our in group nevertheless we support and we externalize their wrong doings by other forces and in that case they would not be punished the way you had predicted and we have to do analysis like this. So, when I did all the simultaneous equation modeling the different things I have written on that side you would see that we had separation in the first graph we had mediation in the second graph we had mediation in the third graph in other that means the original non-significant effect was further reduced in the second and third cases, but it was increased in the first case. So, anything that increases the effect is a suppressing variable. So, if you look at the three structural equation modeling here in the first case non-significant effect is being made significant if you bring in outrage in the other two cases the non-significant interaction regression coefficient that is reduced. So, that means mediation is taking place here and when you come to the right side you see like I am giving you direct effect and indirect effect like point three zero and the confidence interval is between point zero two to one zero one. So, that one is greater than zero and in the second case I say that it is one point one one and the confidence interval is between point zero two six to two two four. So, that one is also a significant mediation. So, your overall non-significant interaction was because of suppression and mediation mediating variables if you control them picture is clearer. So, be a good not a bad servant of statistics. Now, so you see like both outrage was a suppressor external attribution and attitude bear mediators. So, what the points I am making here emotions attributions and attitudes like the defending discrimination by your in group leader do play causal that is suppressing and mediating roles in persecution of the leaders accused of gender discrimination in organizations to tease out these causal roles. However, it is necessary to specify which whether an intervening variable is a suppressor or it is a mediator and which one is a proximal and which one is a distal variable means who which one should be closer to the IV and which one should be closer to the DV. Now, I give you a another fourth example for this rare effect what do I mean by rare effect here one example I said in that I am management review 2012. We manipulated two things circumstances of crime somebody committed crime intentionally or he was provoked extenuating circumstances and the victim had a major consequence or a minor consequence high and low plus we had participants from two cultures United States and Singapore. So, we have really three kinds of things culture circumstances and severity. And what we measured we ask the participant here is a newspaper report in which a lady has been hurt under these circumstances one of the two and this is the severity. So, I want you to say whether this fellow is this kind of person this positional attribution whoever committed perpetrator is a criminal type. Second one how much you would point finger at him means blame him moral responsibility. And third one how long he should be sent to jail three things I am measuring here. So, our design is essentially a three way two into two purely between participants factorial look at these means now. So, three responses I am showing in my three rows imprisonment blame and dispositional attribution. And the three things circumstances severity and culture I am showing in the columns with two levels. Let us look at imprisonment on imprisonment you have effect of only severity of consequence. So, whether crime is committed intentionally or under extenuating circumstances we say same level of punishment. Whether we give this to American or we give to Singaporean they give the same level of punishment. The only effect you find that those who had committed crime of severe consequence is being punished more 5.88. Then one who had committed crime of low severity that is 4.52 it is significant difference got it. So, your dependent variable has one main effect not the two other main effects. So, should a law say that there should be no mitigating circumstances while punishing a person this would be implication for law. Then come to the second one moral judgment now. In moral judgment you see there is effect of circumstances. The person was blamed more when he had committed circumstance intentionally than when he was provoked by his friend to do. When it came to severity there is no effect on severity on blame, but there is a cultural difference. American blamed that perpetrator more than Singaporeans. So, when we come to blame there are two effects when we come to punishment one effect. When we come to attribution let us come to the third variable you have effect of circumstances on attribution. Then we have no effect of severity and we have no effect of culture. How to make a coherent test story out of this data? One would spend whole life writing the story about it and you would be making wrong recommendations based on this data. If you have not conceptualized it properly. So, first I have shown you the findings because my purpose is to show how to make sense out of non significant findings. Let us come to the model which we were testing. I had proposed a causal moral model of imprisonment. Now in this model we say that some of these variables influence your dv and mv through different routes. For example, the effect these effects may be mediated by causal attribution that is like so and so this kind of person. It may be mediated by moral not causal attribution. They may have a direct effect on punishment. So, if you put like this and my hypothesis there are 12 authors of this paper from so many countries because we had a large program on this. So, our hypothesis is severity of outcome is an example of a direct variable greater the severity greater the punishment automatic tendency we have. Then we have a second culture is an example of a variable mediated by moral responsibility. So, like Richard Nesbeth says cultures differ in attribution I say no they do not differ in attribution they differ in assigning moral responsibility. This is what I am saying American and Singaporeans would differ in assigning moral responsibility not attribution and that you have seen. And third thing I am saying circumstances affect this positional attribution. So, when you have done intentionally people would say you are more criminal than when you were provoked to do. So, if you put in this story and look at this chart a story becomes very simple to write and simple to understand So, all I have done here circumstances leads to dispositional attribution, culture leads to blame, but blame is being built by dispositional attribution and culture. Just like the actually this study gave me idea for the sequential mediation. You see here situation is leading you to make causal attribution, disposition. So, that plus culture the cultural notions which my colleagues have been talking here the two are building the blame. And once the blame is built now you see punishment is being determined by severity of outcome and also throw blame. So, both causal attribution blame determine imprisonment and this is what the legal system says. So, things which appeared very difficult to understand if you put within a model and analyze correctly the non-significant effects on punishment. They are not as such circumstances have effect on punishment, culture has effect on punishments. We have to conceptualize them appropriately and this is what I have done here. So, once we do it and look at here when we tested this model look at the fit indices chi s square is non-significant n n f i is 1.04, i f i is 1.01, r m s e is 0 0, s r m r is 0.04, a satisfactory fit to the model and all paths are significant. So, a story can be made simple like this here, but if I reverse some of the variables in the same model you see they did not fit to the data that is the correct conceptualization of the sequence. If you alternate them then they fail this is what I am demonstrating you at the bottom. So, we have to here we are learning effectively then 4 points here. The effect of coefficient you see here proximal variable of circumstances on disposition you see regression weight is 1.49. So, any proximal effect is a stronger effect of circumstances on dispositional attribution is 0.149 which is a strong one. So, another effect of culture on blame is 0.87, effect of severity is 1.41, but when you compare the mediators their effects are 0.036 and 0.043. That means, we have to compare we also need to understand proximal variable and distal variables in any causal chain which one is closer to the i b which one is closer to the d v and this has implication for understanding. So, distal variable have less effect on the d v like circumstances and blame and because those effects are really observed by the distal variables which were occurring at the initial stage and when we come to. So, through ascertain this effect you have to consider the full causal chain not part of it. As Tindall suggested what came what would follow what would happen this is the way we have to do it and that idea we are getting from chemistry mathematics and physics clouds we were ice they are suggesting how mental processes would be a study. And if you put like this to sequentially dependent a weaker effect of i b and d v we have to use the simultaneous equation modeling that is the correct analysis using t test regression analysis is not the correct way. So, let us go back to the public policy issue that if psychology is to become relevant for India how this issue we could have explained to the government very simple answer we have given. So, many intervening agencies some suppressor some mediator some distal some proximal a state capital district block panchayat et c between central budget and the beneficiary their location and beneficiary link has to be weak 10 paisa 15 paisa you cannot expect them to be one to one correspondence right and this is what our leaders have been complaining. And what the solution we will recommend yes the i b has to be made proximal to the d v like the government policy saying apka paisa apke haath if we can really implement if internet is working if banking system is working it should be possible to reduce it like through the unique identification number. If there are not so many intervening agencies one rupee would reach as one rupee to the beneficiary and this is what I am demonstrating through these experiments. And the question is why it should happen the answer is very simple that we have to have i b proximal to the d v if there are so many mediating variable we do not know where the excitement where the effect would be lost. So, I had started with a gloomy lady face that my effects are not significant I am showing my own picture old man here now you see he is laughing deliberately I have designed like this. So, what we have learned out of it we can now reduce the effect of both the statistically significant and null total effect of an i b a known event into its direct and indirect effect like a too implicit you know components on the d v which is another known event. Number two a null result is not necessarily of no value insignificant or invalid with proper consideration of the theories and method it is now possible to trace the excitatory and inhibitory forces intervening between the cause and effect regardless of our discipline whether we are in economic psychology biology head call sciences animal husbandry approach is the same. If we follow the approach we would be able to answer it and thank you very much for your attention.