Behavioral tests in general not very pricious. All animals are a bit different from each other. And they are very sensitive to enviornmental conditions. Even that are not sufficient to people. So it's necessary to use large amount of animals or a lot of trials with each animal. And also it's necessary to do it in same conditions, so behavioral lab should be well conditioned, it should be have same temperature, same humidity, same preassure etc.Lab equipment also should be checked and adjusted regulary.

And also it's not very good to use female animals for behavioral tests due to changing of their behavior during estrus cycle. Try to use only male.

What are possible reasons for high standard deviation in mice experiments?

There are a lot of reasons. I agree with Dr. Brendan M Walker and Dr. Viacheslav Viatchenko-Karpinski.

When you say "different mice", do you mean mice of the same strain, or is there a natural difference between the groups?

As already pointed out, there are a range of different reasons why behaviour would fluctuate between different mice. We had an experiment where some rats showed reduced behavioural activity for no apparent reason, and we later found out it was because the cleaners in the hallway slammed doors and listened to music, which distracted the rats during this particular testing phase.

It is important to remember that mice, like humans, are individuals. They will not show exactly the same behaviour even if all variables are controlled for. I am more acquainted with rats than mice, but I can tell you that rats will show a lot of variation between individuals for no apparent reason. An example is a study from Kansas State University where rats had to choose between a small, immediate reinforcer and a large, delayed reinforcer. It was predicted that the rats would chose the larger reinforcer, as it had more value. Amazingly, some rats chose the large reinforcer only 1% of the time, while others 92% of the time, despite being the same age, strain, and being in the same conditions. In our own lab, we sometimes see "Einstein rats", which learn a behaviour very quickly, while the rest learn it slowly, even though the homecage and experimental setup is carefully controlled. This inflates your standard deviation, and thus some researchers tend to split their sample into two subgroups that perform high and low, in order to control for this variation (e.g. Adriani et al., 2003/2004).

That your mice show variation in their performance is difficult to avoid, however you can control for it when you do your analysis. I don't know what your hypothesis is, but if you are comparing performance between a baseline and a manipulation, then what matters is the difference between the conditions. If there is variability in both groups, then you will still be able to extract a difference between the conditions. If you are using a within-subject ANOVA, then this is arguably even better, because it can control for variance within the same subject. Of course, as your standard deviation increases, the chance of a significant result decreases. If your standard deviation is very high, it suggests that behavioural variability was high, perhaps due to an unknown third variable which you have yet to discover.

As  Dr. Walker and Dr. Viatchenko-Karpinski has already stated, the only way to experimentally control variation is to ensure that all the mice are experiencing the exact same conditions. They should be the same age, strain, gender etc. They should be trained and handled the same way. They should be tested the same time of day, fed the same amount of food, have the same amount of cagemates, etc. The more variables you control, the more validity your experiment gains. But even then you should still expect some variation between the individual mice.

All the above points are well made, except one. Females are not on average more variable than males looking across many traits. Please see the below article on this point. 

It is possible that females are different from males in the particular behavior(s) you are interested in, but you have to show this and not assume it. 

http://www.ncbi.nlm.nih.gov/pubmed/24456941

Thank you very much for the responses and the help, Drs. Brozek, Sjoberg, Zabrodskii, Walker, Viatchenko-Karpinski. Your input is extremely important and I hope it will help me in improving my ability to get cleaner results.

Dr. Sjoberg, the mice were of same genotype, age and sex. The only difference between the mice is the treatment. 

Dr. Walker, the room we perform our behavioral analyses is pretty isolated and mostly calm. The mice were handled intensively but all the groups were handled uniformly. 

Thanks again for the input.

If you are dealing with mice and anxiety-sensitive tests big (and sometimes unpredictable) variance is somewhat to be expected.

Only after n=15 (for inbred B6 mice) you can really rely on the results. Smaller cohorts can also provide insight of course, but unless your effect is very robust it is very easy to make false assumptions from these and this can cost you a lot of time later on. This is very common if you analyze many behavioral parameters, some will give positive outcome based on chance alone.

Mice live in community (cage) and every individual is perceiving day-to-day reality in a different way. If you calculate variation in published results you will very often find huge variance in some experimental groups.

One of the solutions is to single-cage the animals two weeks before the behavioral test. This decreases the variance, but it includes single-cage-stress component which can easily confound the outcome.

One has to balance, and it is not easy. You need to know what is the effect size you want to evaluate, if it is 30%+ than using n=10 is OK assuming that the variances are not huge (StDev/Mean < 25%) for initial experiment . If the effect size is lower than 15%, unless it is something you really really need in your paper I would not even do the behavioral experiment.

There are no hard rules and guidelines I'm afraid, this is just my rule of a thumb

Hello! The great variability in behavioral tests is a huge problem, but at the same time is a fantastic area to explore. What I mean is the studying of individual differences. Therefore, I would recommend not only to think about the ways of controlling / removing the variance but about  its origin as well. Studies on individual differences may still contribute significantly our knowledge about the universal phenomena.

Best,

W

Hi, most points are already mentioned above. Additionally, we gained results depending on the person handling and conducting the trials. Even though all should follow the same procedure, different persons get used to different ways of handling animals and produce great variation.  There is a nice open source paper of Eccard and Herde showing that wild voles are consistently bold or shy over liefetime and different "personality" tests.

Herde A & Eccard JA. (2013) Consistency in boldness, activity and exploration at different stages of life. BMC Ecology 13:49 doi:10.1186/1472-6785-13-49

cheers, Thilo

You didn't mention the specific test or details of how you're doing it, but In addition to the things suggested by Brendan, make sure that the timing of your treatment and control animals is the same.  That is, for any day of testing, make sure there are control and treatment animals being tested.  If you are running squads of animals throughout the day, make sure that each squad has approximately the same number of control and treatment animals so if there are any influences of time of day that are distributed among both groups (and you can test for a time of day influence).

My apologies for not mentioning the tests I'm performing. I'm performing Open field, Light-Dark box, Novel object recognition, Y maze and water maze.

Thank you very much for all the suggestions and replies. 

@ Viacheslav Viatchenko-Karpinski

You mention that " it's not very good to use female animals for behavioral tests due to changing of their behavior during estrus cycle". You would then assume that any behaviour elicited by male mice/rats would then be generalized to both sexes (one should not assume that a female brain, abundant in estrogen receptors would elicit similar results as a male brain). That in itself is a narrow view, which is likely to create confusion. I do agree to a degree (depending on the the behavioural test used) that males and females should be separated, and I would add that female subjects should always be included in every study, given the fact that pretty much any mental disorder (from symptom to treatment) seems to be gender-dependent.

@ Goutham Kodakandla 

As someone mentioned earlier, make sure your apparatus is clean, or at least the potential odours are equally spread throughout the apparatus, especially for the NOP test. if you DO end up using females, make sure you control for hormonal status either via OVX or by recording the estrous phase on a daily basis (don't assume that they cycle regularly, because they don't). In addition, females in high estrogen states tend to be faster, and that can affect your outcome if you are measuring any latencies.

skewness is the usual suspect. you may calculate your "deviations" with a new approach

Article TO DETERMINE SKEWNESS, MEAN AND DEVIATION WITH A NEW APPROAC...

The animal groups used have to be small. In a group of 10 only one animal needs to deviate from the other e.g. in maze learning and your SD is big.