Nasir Mustafa Teşekkür ederim hocam. Ben de tüm deneme periyotlarının ortalamasının alınmasına dair protokolü bu kaynağı referans alarak uygulamıştım. Ancak, deneme periyotları içerisinde herhangi bir trialin tek başına kullanılıp kullanılamayacağına dair bir bilginin veya uygulamanın varlığından emin olamıyorum. Gözden kaçırdığım bir detay varsa öğrenmek isterim. Teşekkürler.

Thank you for your help. I also analyzed my data via averaging four trials per day according to my reference that you've also shared. However, I am not sure whether there is any information or practice regarding using any of the four trials (e.g. first trial of every acquisition day) instead of the average of four. Please let me know if anything is missed. Thank you.

A couple of my mentors/colleagues are 'hard liners' that actually trained under Dr. Morris. They would say plot each trial on each day. Most of us, as you say, use block (or day) averages instead, and treat each animal's day average as a single measure, and then use a repeated measures ANOVA to statistically test.

My reasoning is threefold: 1- we're starting the animal from different quadrants on each trial- that means that for the two goal-adjacent quadrants, the direct path-to-goal is shorter than the direct-path-to-goal for the 'goal-cross-quadrant' starting position. Since we deliberately do not use the same quadrant drop order every training day, each trial's drop quadrant does not align across training days. 2- there is variability in the animal performance that is not associated with learning. 3- the amount of learning appears to be greater between days than it is between trials on the same day. So, block/day averaging does remove the within-day trial learning , but it also controls for non-learning-related biological variability as well as the quadrant drop effect (as long as all quadrant drop locations have equal representation on each training day).

The first trial on each day sometimes gets special attention. For instance, some papers only report the first trial of each day, while others report the day average but exclude the first trial from that average. The rationale behind treating the first trial as 'special' is that some animals (in my experience, this is more often the case with mice than rats) are more distressed during the first trial. You can see this by investigating the path traces for thigmotaxis (interpreted here as circling around the edge of the pool).

You may also see some researchers drop their last trial per day, reasoning that the animals are becoming exhausted- if you are consistently seeing longer latency/ path length on your last trial, you may consider modifying your design- either removing the 4th trial from analysis or increasing the inter-trial interval for future experiments. Water temperature (we keep ours at 26 degrees C) can also be an issue, with colder water fatiguing the animals. Again, this happens more easily with mice than rats.

Now, all of these approaches are appropriate in certain conditions, and obviously are 'allowed' because they have been published- a researcher just has to state what they did and their reasoning for it in the Methods section, and then some reviewers and an editor have to agree. But, this does create a temptation. Is the researcher cleaning up the data to make their results more clear, or is the researcher 'sculpting' the data to fit their hypothesis? You'll know you're doing it wrong if you are watching the output of your filtering strategies to find the graph that best supports your hypothesis.

Finding a difference in the first day between groups could mean there is a difference in mobility between control and experimental groups. The swim velocity during visual cue trials may help to determine this. If it is the case, consider using path length, or Gallagher's spatial learning index, instead of latency. This may be especially important for experimental conditions associated with Parkinson's or aging, where movement is a concern.

Am ok with Eric M Blalock

In my own case, I used daily average of all trials from day 1 to day 5, and analysed them through two way ANOVA with Tukey post hoc. For the graph, I used time-line which gave me a beautiful learning curve.

Thank you.

Dear Dr. Eric M Blalock

Thank you very much for this detailed and informative explanation. You generously addressed the guiding principles of Morris Water Maze to a researcher looking for that kind of information with reasons.

I tried to analyze my data in every possible way. Firstly, I plotted the first trials of each day. Then, I used day averages. When I compare those two different analyses, I can see only slight differences between the results. That makes me think that I am on the right way (Hope I am).

I have a question about your reasoning. You told that ‘’each trial's drop quadrant does not align across training days since we do not use the same quadrant drop order every training day’'. Doesn’t it mean that we may prevent misinterpreting our data by using day averages instead of a certain trial on each day because we take account every different drop quadrants equally in day averages?

Since I’ve too many animals to test (n=20-25 for each group), it was very difficult to handle with their heterogeneous behaviors. To avoid bias, I used the Interquartile Range (IQR) rule to find my outliers and removed a few outliers from my data set and analyzed with repeated measures ANOVA than Tukey post hoc. As you told, I found a difference on the first day between groups. It may be because of the difference in mobility. When I analyzed the swimming velocity, there was a significant difference only in the first day, not in the rest of the acquisition trials or probe trial. Then, I calculated the path lengths and it showed a similar pattern with escape latency. These results show that we can interpret learning and memory functions independently of motor functions, can’t we?

I am grateful to you for this informative discussion. Your words helped me a lot!

Best Regards.

Dear Dr. Musa Samaila Chiroma

Thank you for your comment. I analyzed my data by using the first trials and also by using day averages and determined only slight differences between the two methods. Although most of the papers mention using day averages, I had some concerns about it. Dear Dr. Eric M Blalock helped me to clarify my concerns.

I used two way ANOVA then Tukey post hoc test to create my learning graph. Thanks for sharing your experiences.

Hi Dilara Nemutlu Samur,

Thanks, when I think of how complex data analysis can be for 'watching a rat swim in a pool of water' I marvel that we expect to make sense of the universe.

I was taught that the quadrant drops were balanced across training days because animals could learn the drop sequence and then use what we call 'dead reckoning' (from the rat's perspective- if you drop me in this quadrant, I always turn left- if you drop me in that quadrant, I always swim straight across)- it's a form of learning but it's not the one we're trying to measure. But yes, it probably means if trial 1 on day 1 was started in drop quadrant 1, and trial 1 on day 2 was started in drop quadrant 2, then comparing trial 1 on day 1 to trial 1 on day 2 has an issue.

I've seen quartiles or 2SD (or even 1.5 SD) used to identify (and treat as missing data) outlier variables. Make sure that the missing data is tracked in the n's for statistical analysis (statistical power is lost as values are dropped from the data table). It usually doesn't help as much for day 1 measures since so many animals time out, leading to non-normally distributed data (some researchers prefer to analyze the ranked data because of this).

I don't think that the latencies and path lengths can be said to be independent of one another. If you ran a Pearson's correlation, you would probably see that they are very highly correlated. Water maze depends on movement, so there probably is no good measure of cognition in water maze that is not in some way potentially impacted by differences in movement between groups. However, path length is less dependent on swim speed, so you could say "to guard against differences in swim speed between the two groups" when you report it.

For the swim speed, it would be best to use data from visual cue trials (if they were run). For me, the problem with using the spatial cue trials to estimate swim speed is best summarized by an analogy. Say that you park your car in a parking lot next to a market. You go in, shop for a while, and then come out. Now, do you walk faster if you remember where you parked your car, or if you don't remember where you parked your car? I think the rats are probably a little slower when they don't remember the platform location. So it would be safer to use the visual cue task, when they know exactly where the platform is, and swim straight to it.

If your data supports a significant difference in swim speed between the two groups, you can say this in the text of the results, but also mention that both groups learned. Then, for the probe trial, use the path length in goal area (or goal quadrant) for your main measure. If this is an animal model you will work with more, and you will have n's of 20-25, you could try using novel object recognition. Novel object recognition needs movement to measure, but less so than water maze. As a caveat, novel object is not considered as sensitive as water maze at picking up cognitive differences

Dear Dr. Eric M Blalock ,

Thank you for your precious time on this topic. You gave me the clear key points of MWM and it really helped me to use my data and interpret them correctly.

Actually, I did not track the missing data in my ‘’n’’ values since I thought there could be some ethical issues with it. That is probably due to my lack of experience. When I tracked the missing data in the ‘’n’’, the significant effect was the same but the statistical powers were different as expected. I had some concerns about it, thank you for clarifying.

I fell in love with your style to explain the rat behaviors. It is catchy and easy to remember. Now, the basic aspects of the test are more apparent and interpretable for me. When I analyzed swim speed, there was no difference between experimental groups in probe trial, however, I followed your advice and analyzed the path length in the target quadrant as well as escape latency. They were highly correlated.

When I read your recommendation on using the novel object recognition test, I felt very happy since I’ve already used this test in my experiment. I wanted to support my cognitive data with different tests and I performed novel object recognition along with the MWM. I had to exclude some animals with very poor motor function since they stayed stable when I put them in the box. However, I can see that the NOR data correlates with the MWM probe.

Your comments added a lot to me. It was great to discuss with you. Thank you for being an ‘’online mentor’’ 😊

Kind Regards.

Glad I could help, good luck out there!

A way of assessing the acquisition learning is to average escape latencies for each animal (for 4 trials) for one trial day. Then measure mean + SD for each group and compare by repeated measures one way ANOVA over 5 days, followed by multiple comparisons. Shallow curve is representative of learning deficit.

A quick question, do you think Parkinson rats/mice will anyways show a deficit on acquisition trials given that motor coordination will be compromised.

In this case total distance covered will need to be factored in.