you could use ANOVA, or time series

for SPSS Try this link : http://www.ats.ucla.edu/stat/spss/topics/anova.htm

or youtube: http://www.youtube.com/watch?v=9TDvejlOP0Q

for Minitab

http://www.youtube.com/watch?v=ymSIjaygcVM

https://onlinecourses.science.psu.edu/stat502/node/66

Thank you, I have no problem on using ANOVA but my question is perhaphs more specifical - how can I compare different site of the same species (oak) but exhibitng different quantitative seed produciton (tons/ha). Should I normalize all the series and therefore using parametric stats or I have to use rank distribution by means of percentile and therefore using non-parametric stats? Thanks

Hi Francesco,

Do you have observations from multiple trees at each site, and any idea of the time-course of seed production per tree in each year? I'm trying to work out if your data are predominantly snapshots of seed biomass for each season, or if you have details of when seeds are produced within each season? The options for analysis will be different depending on the detail of the data.

Hi Neil, I used littertraps (9-15 per plot) to collect litter during fall. Litter is collected at least four times with the last collection timed soon after last litterfall. Therefore, I could have idea of the time-course of seed production at plot level...However, owing to a long time series I have (15-20 ys) and several sites (~10) my idea is to find annual synchrony rather seasonal patterns...

Hi Francesco, when you say "annual synchrony" do you mean the extent to which mast is produced at the same time in different locations in each season, or the extent to which, across all locations, mast production occurs at the same time each season? Sorry to ask so many questions - I just want to be clear what it is that you are interested to know,

Uhm I believe the extent to which across all location mast occurs at the same years. For argument sake I would test if oak mast occurs at the same time regardless of site and stand conditions (i.e. mountainnous and hill environments) or for example if stands located in hill areas showed a lag of.one year.compared with mast occured in a mountaionous areas

Hi Francesco,

From what you say, it seems a "time-to-event" (or survival) model will be appropriate for considering individual mast events. If, over your time series of observations there are multiple mast events in each location it may be more appropriate to examine the time series properties of each location or type of location. In this case you could examine the periodicity of events as a multi-event indicator that is similar to synchronicity, I think.

Thank you very much for your help. I will looking for these suggestions! The sites I have have about 4-7 mast events. I will check for these statistiocal stuff's guidelines to try to perform my analysis. thank you.

Hope it works out for you. Yours,

N

dear friend Francesco, You have applied ANOVA with multiple compassion like SNK, Tukeys test, It is very much useful for data processing, The above concern test was separate the location with mean value of ur data, Best of luck

Thanks all

Hello Francesco, 10-20 years of data is nicely unusual, and may have interesting things to say. I've (quickly) read the responses and your responses, and I'm still not absolutely sure what you are trying to do because your description is quite brief.

I would suggest that you write your best shot at a summary-research-proposal, outlining the data you will be using and the question you want to address, and that way you will limit answers to your true question. That is, unless you already got the answers you needed.

That said, my best guess is that you want to know whether there is consistency across sites in some variable (seed production) in the long term (annual scale). I agree that if you have only four collections per year you will not get any great precision on the seasonal scale unless those four collection times vary randomly from year to year (in which case I would recommend periodic regression; see my book 'Analysing Cycles...').

More generally, I would firstly recommend against any qualitative variables being constructed out of essentially continuous ones --- i.e. don't try analysing for seasonality with ANOVA on 'spring, summer, fall, winter' or on month names, etc., because ANOVA will merely say 'some of these things are not like the others' and will not give you a very useful quantitative model, whereas a periodic regression can describe the cycle (tidal, daily, annual, etc.) in terms of a sinusoidal curve with an amplitude and an acrophase (where it peaks, or, if you have multiple cycles then for each cycle where that cycle's maximum contribution to Y occurs).

Then I would recommend using a 2-stage process, consisting (firstly) of a de-trending process to create anomalies, data from which all easily explainable variation has been removed, and then (secondly) exploration of the anomalies using various plots and time-series methods to detect patterns.

For de-trending, the first stage, you will need to consider what kinds (sources) of variation you wish to retain in the anomalies. The sensible way to begin is to consider all expectable variation, and then decide which sources you do not want to remove (thus that you want to leave in the anomalies).

For the de-trending process, construct a mental model of what you think could (not 'is') drive the phenomenon of interest, and include all the terms that are reasonable (and if you lack data for the most reasonable terms, that is probably the end of it). Then convert your mental model to a regression general-linear-model (like Y = B0 + B1*V1 + B2*V2 + ... Bn*Vn + error) where Bs are coefficients and Vs are variables (like altitude, soil moisture, etc.). You will likely want a 'continuous time' term there too, like sequential days since Jan 0 (when formatting time properly, there must be a zero! ... but it means the first infinitesimally small instant of a year) of the year of the first observation, but you may also choose to omit this to let this long-term variation remain in the anomalies; just think about it carefully. There will be seasonal components to your data (as you have described it) so you will need periodic terms in there (the sine and cosine components of time in year, and you may find a considerably better fit by adding the second harmonics also), and I describe how to do that in 'Analysing Cycles...". These will be required to minimise the variation unexplained by the regression; if you don't address seasonality properly, it will remain as a component of 'anomalies' (below) and will result in a waste of your time; but they are very very easy to include (see the book).

If you have variables that are important and also unavoidably quantitative (e.g. if 'soil type' produces effects in tree growth that is not decomposable into continuous-variable 'soil type' descriptors like contents of sand, silt, etc., so that you are 'stuck' with using 'soil type') you can include them in coded form (though many packages will handle them for you), but always look for continuous variables by preference. The error will be tested for conformance to the distribution assumed in the procedure, but unless you find a compelling reason to say otherwise you'll assume the Normal distribution and provided your errors conform to that you can proceed; if they do not then you might try various transforms of Y and see how that goes.

Once you have a model that gives results that meet the assumptions of the model, you can treat the errors (residuals) as anomalies. Do be sure you are happy with the model though, because that was probably the easy part. The forest stands or thickets or units you have sampled now each have, for each time you sampled, a residual, or an anomaly (departure from the model). Of course, you might be directly interested in the coefficients (Bs) estimated for each variable, but here from what I read you were interested in year-to-year similarity or pattern. That is why I suspect you'll do better with data that have been 'cleaned' of everything that is already obvious enough for you to have included it in the regression model.

Your next problem is how to analyse those anomalies. I would begin by exploring them visually. E.g. by plotting them, as traces for each species, or for each forest block sampled, etc. You may during this process realise that you need a different kind of de-trending analysis (what I described above), and generate residuals (anomalies) from a different model, and come back to this point with those. Once you have a visual understanding, you will most likely proceed to time-series analysis of some kind.

I'll end here, because going further may be going in a direction you do not want. If you want to go into more detail than is practical here, e.g. if you need to communicate with diagrams, you may contact me directly. Best of luck, --- Kim

Dear Kim,

First of all I thank you for the insightful comments.

I have to go in depth with your comments step by step, and If I will have a problem I will ask you. In the meanwhile, Thanks a lot!