Don't try to find how you can force your data into a given tool. Better find what tool might be appropriate for your data.

You should use a multilevel model using sample ID as random factor.

It may also be good to find a sensible functional relationship between time and response (what is called regression, not Anova, because the main predictor is quantitative and not categorical).

See:

http://www.stat.columbia.edu/~gelman/arm/

Don't try to find how you can force your data into a given tool. Better find what tool might be appropriate for your data.

You should use a multilevel model using sample ID as random factor.

It may also be good to find a sensible functional relationship between time and response (what is called regression, not Anova, because the main predictor is quantitative and not categorical).

See:

http://www.stat.columbia.edu/~gelman/arm/

Hi,

if you are looking for significative differences in the mean value of the tree series (if the series have the same mean value or not) you can perform an "ANOVA type" analysis using the timeseries data as statistical samples but you have to account for series autocorrelation which havily biases the results. Look for "ANOVA for autocorrelated samples".

if you are looking for significative differences in the mean value of the tree series;  you can perform an "ANOVA type" analysis using the time series data as statistical samples but you have to account for series autocorrelation which havily biases the results. 

It may also be good to find a sensible functional relationship between time and response using regression methods.

I suggest you treat your repeated measures (time series) as multivariate observations. Then choose the design (multivariate) as appropriate.

Dear Odai Judeh

I agree with most researchers answer, and I think these links useful 

Best Wishes,

stat.smmu.edu.cn/DOWNLOAD/ebook/statistics_course.pdf

www.stat.columbia.edu/~martin/W2024/R3.pdf

www.stat.cmu.edu/~hseltman/309/Book/chapter7.pdf

www.uvm.edu/~dhowell/fundamentals/SPSSManual/.../SPSSChapter8.pdf

https://wikieducator.org/images/f/fb/Stats_11_ANOVA.pdf

You should discribe the type of your outcome and how many days you observed.

You need a GLM or mixed model because time-dependent data; using of classical ANOVA is forbidden in a situation like this. Also, only long time series data are suitable for the longitudinal analysis. You can look at https://support.sas.com/documentation/cdl/en/statug/63033/HTML/default/viewer.htm#statug_glm_sect036.htm for some clarification of repeated measures analysis of variance. Before analysis, you should check the data with Bartlett's test of sphericity test and Levene's test. The time effect with a small number of categories (time events) should be treated as fixed effects.

Good luck with analysis

One-way anova implies that there are no day effects. Days do not constitute a source of variation. That is, you are not blocking the experiment by days. If the conditions from day to day are fairly uniform you can consider the experiment as comparison of three samples. One way ANOVA can then be used to compare the three samples.

All the the above contributions are totaly true, but I do not want to complicate things for you. If you want to compare the means among the three samples and you sure that these samples are independent, you should check the homogeneuity of the three samples and the normality assumptions, I think you may use the One-Way ANOVA test using SPSS or Minitab or any other tools. In case of violation of the normality assumptions, you may use the Kruskal-Walis H test using SPSS too. For more informations or need help in using SPSS you can contact me on my e-mail: [email protected]

Dear Basel,

that is exactly hitting the point: the data of a time series are not independent. If, for instance, the values at a given time point are (very) high, the values of the adjacent time points will also be likely higher* - and this means "dependence".

That's another reason - I think - why ANOVA does not make any (well, at least not much) sense here.

* unless the selected times are not so far from each other that any connection is lost... but tha't wouldn't be a time course analysis, anyway

What I mean if the three variable are independent for example to compare betwee three countries, three universities, three jobs, etc. if this is okey you do not have to wory much on the data for one series. In this case I can use ANOVA. It is different than modeling a one time series data, you have to check carefuly for the components, trend, seasonal, etc.