Hello Manon, in SPSS, you would most likely use Repeated Measures ANOVA with the body weight of each day as the within subject factor and the treatment as the between subject factor. Look out for significant interactions treatment x time. With 7 days and 6 treaments, you need quite a large dataset! If you can use R, it might be better to switch to Mixed Models with the lme4 package. Best regards, Wolfgang

Hello,

As Mr. Wolfgang the ANOVA will be the best choice

You have a factorial Between x Within subjects design,

with SPSS see:

https://www.youtube.com/watch?v=m18XZ6YUkG4

This is a complicated question. Undoubtedly one of the worlds experts on these kinds on models is Prof. Marie Davidian. Her revised course notes are Here:

https://www4.stat.ncsu.edu/~davidian/st732/notes.html

Best wishes, David Booth

Manon Eggink

I hope this isn't a curve ball but many time-based events can benefit from a graphical or variogram approach.

I may just know enough about statistical analyses to realise how important statistics are in the meaningful analysis of any data-set and to get an expert involved. However, in my experience, experts often suggest approaches that are not practical neither achievable in real life. It follows that the onus is on the researcher to have enough knowledge about a subject and the practical experience and confidence to argue the point with the expert. Over the years I have found that eventually they may understand what you need and recommend appropriate/applicable alternatives (statistical analytical tools) that will enable you to determine (in a practical manner) the built in error and the associated risks in the approach that you follow. Of great importance is that the researcher need to understand (and fully evaluate) the various "meaningful" factors that could influence a dataset and to measure all on it as part of the dataset. without this understanding no statistical analysis of a data-set will be meaningful.

I think you should use Repeated Measures ANOVA.

• Gerrit Jordaan Perhaps in some cases that's true. The person I suggested developed the methods and used them in series of studies on the success of drug cocktail treatments for Aids which are required because of the rapid mutation of the HIV virus. That speed made the use of Repeated measures ANOVA methods "impractical". Perhaps you've forgotten that R. A. Fisher was invited to India to found the Indian Statistical Institute because the Government of India cared about the number of starving people and wished to improve their agricultural output , Which by the way the ISI did under P. C. Mahalanobis' direction. Mahalanobis was an early student of Fisher. Perhaps a one semester introduction to statistics is not suitable for the research worker any more. Isn't that why we get graduate degrees? Best wishes, David Booth BTW I was involved in flexible pavement too with The Illinois Division of Highways Bureau of R & D. Good luck in your work.

Dear David Eugene Booth, I have not forgotten any thing or trying to belittle the expertise, I was actually emphasising the the need and importance to (firstly) involve an expert statistician involved with the analyses of any data-set. My apologies if it did not come over as such. Secondly, I tried to emphasise the need that the onus is on the researcher to ensure that the expert statistician must also understand the problem and the need for a practical solution. The onus is on the the researcher to make sure that the expert statistician fully understands the nature of the subject and the solution required. Not only should a sound analysis of the data be recommended, but also sound procedures for future use to ensure that data collection is statistically meaningful.

This is a general problem that exists in engineering especially when dealing with materials that are highly variable, where the application of sound statistical analysis procedures are more often than not totally ignored. In such an environment it is even more important to get an expert statistician involved. Since you have done some work with Illinios Highways Bureau, you may have an understanding of some of the frustrations I am dealing with. In transportation engineering, traffic engineers understands and uses statistical procedures all the time. However, in road pavement engineering engineers deal with materials with tremendous variation in material properties and road (pavement) conditions, requiring sound statistical procedures in the analysis of data-sets. In practice I have found this often lacking both in the gathering of the data as well as the analysis of the data. Hence, I am using this as an example (totally removed from the human factors you have mentioned).

Most of the road authorities around the world are using tests and data indicative of structural evaluations, the frequency of frequency of which will immediately result in a +- error of more than 50 % (without realising that it is the case). This data is then used in analytical procedures in detailed design where a +- error of (maybe) 0% is required, because of the financial impact and consequences on the final decision to be taken. I have seen take numerous such case where no cognisance exists of the impact of (for example) the frequency of the data on the accuracy of the analyses to be done (this is happening on a daily basis in the modern era all the time). Many a PhD study have been produced without any idea of the statistical built-in error of their data. In my opinion engineering courses (for example) should as a major subject, have a statistical (practical) course as part of their curriculum throughout the duration of training (not just as a once of, by the way subject) because the essence of engineering is risk management and if you don't understand the impact of your data you cannot (in practice ) optimise the data requirements to achieve the required design.

Many a forensic analysis I have been involved in, investigating failures, can be attributed to design errors due to the fact that no statistically sound testing procedure has been used in the gathering of the information before using the data to do sophisticated detailed analyses. It is a matter of garbage in garbage out.

In these cases the statistical solutions derived at must take into account the practicality of the situation of the process of data gathering, (for example, test laboratories cannot go back and forth to establish initial variations within material as a required input into statistical models before the model can give the required number of tests to be done in order to obtain a data-set with a required +- error - it is just not a practical statistical solution). In other fields the solutions may be much more simple.

Gerrit Jordaan I certainly agree with you but you seem to arguing for greater collaboration in your second reply and less in your first. I was originally a chemist then trained as a statistician. I would argue that each should get out of his comfort zone sometime and i think many statisticians do. I also remark that i don't see any engineers in physics for poets courses but in my 40 + years as a statistics professor I found many more engineers in my diff. eqs classes than I did in statistics. I rarely knew a PE that needed to solve an advanced ODE than needed Stats, at least not in our office. I as a PSTAT am required to give the best advice I can on stat. matters which includes the problem at hand and if I do not request the engineer's advice i would be derelict in my duty. It seems that many other disciplines are not. I am here reminded of the statistician's lament, "Why didn't you call me at the start?". Honestly we are much more useful in design than in rescue. May the Force be with you. D. Booth

David Eugene Booth - 100% in agreement. Unfortunately too many P.E. are just too applying a so-called "safety factor" without any idea of the built-in probability of failure when designing macro infrastructure projects. (Sorry for my "typo": +- error should obviously read 10% not 0%).

In this discussion we have seriously deviated from the original question. However, I just wanted to emphasize that the researcher should also ensure that the statistical solution recommended properly addresses the issue at hand and, as rightly pointed out by yourself, vice versa.

Meet you at the end of the Milky way - Gerrit Jordaan

@Gerrit Jordan Thanks very much for explaining your position. I find that I am pretty much in agreement with everything you said. I certainly agree with you on the idea that the engineer and the statistician need to compliment each other. I honestly think that if I don't understand the problem I have almost zero chance of solving it. That is why most statistics students are advised to study an applied or natural science to get experience in doing that. In closing I am reminded of Spock when he said a difference that makes no difference is no difference. Best wishes, David Booth