Oscar Kempthorne: From Observation to Inference, 1991 part 3

at the national level the shop has trained great man statisticians and you personally I've trained 42 I think PhDs can you tell us about that impact on on statistics in the United States well that great I think it had great influence on biometrics and the biometric society is though you see what's happened is that medical research discovered statistics maybe around nineteen fifty or something so there are hordes of statisticians coming out of the health sciences and the drug firms and so on so you know there isn't as much of the early work yes the shop yes it's it's more being done now by biostatistics departments whereas the shock took took that responsibility in early days yes I think that I'd like to talk now a little about your first book the design and analysis of experiments published by wiley in 1952 that was an enormous amount of work for you and has been a classic has been established as a classic what do you think the message is in that book well that book is really sort of curious because because it presented gave a decent presentation of linear model theory and I think it had some status in that respect yes but then I suppose the big thing that it was pushing was randomization yes coming out of Fisher which and you know there are all sorts of ironies in this because Fisher I believe didn't really understand atomization why is that well might because he says he said well you must randomized but then having randomized you can use gauss-markov normal linear model theory but there's no justification for that and so if you randomize over a finite set yes then you don't simulate in the randomization you don't simulate sampling from a normal distribution which is a continuous distribution yes so you see the it's all very curious well I'm as you know interested in the analysis of spatial data and one thing that I've observed is that he proposed it partly to get rid of spatial correlation oh yeah in fact on page 64 of fishes 1937 book I'll read you a quote Fisher said after choosing the area we usually have no guidance beyond the widely verified fact that patches in close proximity are commonly more alike as judged by the yield of crops the nose which are further apart what do you think of recent renewed interest in spatial aspects of field trials well the this could well be fertile but because there are sort of big difficulties because one has to determine a model of spatial variation and there will not be that won't be a standard thing you know well that requires the skill of the statistician of the modeler working in a team with by attrition I think it's this role of statistics in agriculture over the years that's been very important for the shop and for you do you see other areas where statistics is having or will have a comparable impact well I guess there's no question about medical research is he after all what goes on in medical research is that people are ill so one wants an intervention a pill that will cure it and so the only thing to do is to do a comparative trial okay and you know back around nineteen fifty there was some discussion by Bradford Hill as to whether one could use randomized trials for medical research and you know big argument about it whereas now it's a you know if you if you don't randomized you're an idiot haha can't believe that Kempthorne will blow you out of the whore or whatever this so what about a notion of restricted randomization well such as appears in spatial statistics well the idea that you could have treatments to ensure balance of treatment yeah close to each other yet randomizing within that oh yes well this of course this idea came up in the Latin square design or in randomized blogs the land it's been you know pushed off and on over the years rosemary Bailey in England has been pushing it but these people who are pushing it seems to me don't really understand the game they seem to be very comfortable if you say ISIL to give us like the technical thing now is the expectation of the treatment mean square is equal to expectation of the ER mean square under the null hypothesis I think that's called validity you well they put this up for the validity but you see that is that's a very weak validity yes and you know one surely needs more than that and the thing that one needs you know is randomization tests so distributions rather than expectation please yes yeah and of course if one if one just has two possible randomizations and you use one of them the only statistical test that's available is one that's has size fifty percent yes that's pretty weak thing yes so okay yes is there anything more you'd like to say about your work in experimental design because I'd like to move on to another large portion of your work namely genetic statistics oh yes well we have to move on to Jeanette yeah I think I think we should yeah let me preface this question by saying of Ronald flesh Fisher clearly had a very big influence on your design book and Fisher did a large amount of research in genetics tell us about your own work in statistical genetics and fishes influence on it yes well when I came here is he i came to iowa state iowa state was a leader in the world in animal and plant breeding uh-huh there was an individual he's gone now called lush animal breeding and then there's an individual who's still alive in plant breeding called Sprague and so you know these had students coming here all sorts of students from all over the shop lots of students from Australia graduate students these needed statistics and they needed to understand the role of statistical thinking in the genetic modeling mm-hmm so it was very natural you know that I said now of course I really got involved in genetics way back in the effort to try to understand sure I think because he would call on genetic examples all the time with complete freedom right so if you couldn't understand a bit of genetics you're in bad shape so I think he taught courses here in statistical I taught courses until you know until nineteen sixty i guess and then i was teaching too much yes and so we got we got other people in yes to do some of that yes but so consulted with animal breeders plant breeders yes but I sort of that that stopped somewhat around nineteen sixty because some other people were brought in yes to deal with that no I still go to animal breeding plant breeding seminars yes so well there's a dark side to genetic statistics and that I call it a dark side that's the eugenics move yes now Fisher Ronald Fisher was very involved in the eugenics movement can you explain what the eugenics movement was and is and give us your opinion regarding it well the eugenics movement started I suppose because the quality of nations in particular Great Britain I guess was just going to hell because you know people were coming in from all over the world and they did not have so it was claimed this did not have the intelligence and so on so something should be done about this and Fisher was a big man in this eugenics movement yes now this outlook has sort of come in and out in and out okay yes and it came up again nineteen seventy with an individual called Jensen yes and the he wrote a long thing on the Harvard in the Harvard educational review yes and what did he write well the idea was that that something like you know with very defective thinking that eighty percent of the variation in human abilities mental abilities was due to genetics nice and twenty percent due to environment yes so that's where he came out but of course that's a well that is our own nonsense it's epistemological II can't be sustained they didn't have the data yes you know the only way to establish a causal argument of that sort is by experimentation and you know they these people somehow did not understand have any what for me is a decent understanding of causality you know causality has bugged philosophers or five hundred years I guess longer no longer and no there's a there's a lot of fluff talk in the area the only notion of causation that can be sustained is causation in the sense of human agency that is over in the Ascend solvent of some agent so that's why this experimental design really appealed to me because one does get in a week way for for one does get our paws ality yes and when the doesn't you know one half get it just by sort of looking at data