I have data concerning fruit size, color, mass, soluble solids, and some other phtyochemical properties. Samples are sorted by Year of Harvest, Date of Harvest, Species, and Location. The variables were obtained through widely different techniques of processing, such that I cannot organize a file of raw data that would be coherent. Therefore, I used Excel to calculate all relevant means/standard deviations and pooled these into a single file where each Year/Date/Species/Location Sample is separate but contains the relevant info as a single row in the Excel file
I'm trying to find a way to tell the statistical software (I use R/RCommander, but could use something else if better) that the columns (say, Mean Fruit Mass and standard deviation of Fruit Mass) are means/SDs, not individual data points. I want to be able to conduct Anova analysis, Tukey tests, and Correlation analyses using one document rather than bouncing between 6 or 7 documents. Is this possible? If so, how? Software manuals are being spectacularly unhelpful.
A few things:
First, if you were able to calculate means and SDs for your various samples, I don't see why you couldn't organize the individual observations into one file.
Secondly, doing a correlation would require having the individual observations.
Thirdly, if you must work with the cell means and standard deviations, then I don't think you can use the built-in anova and t-test functions. Instead, why not just plug the numbers into the ANOVA (and other statistics) formula? There's even a simple example here: http://formulas.tutorvista.com/math/anova-formula.html
My issue stems from the fact that, for example, for week 3 in 2015 for species A at location X, I have 150 measurements for fruit perimeter, but 75 for fruit mass, only 3 for Total Phenolics, and 25 for Total Soluble Solids. This is due to the different processing of the samples (75 fruit harvested and massed, photographed on two sides to measure perimeter, a subset removed for soluble solids processing, and the rest pooled and used to measure total phenolics in triplicate). Repeat for 5 species at 4 locations during a 9-week harvest period over 2 years. I'm trying to compare these species to determine what differences (if any) exist between species, between the same species at different locations, between years, and changes over time within the harvest season. I haven't worked with such a large and disparate dataset before. Can I just leave blank rows in Excel and it will work out right? I've been having issues with R on just a subset of the data that has consistent sample numbers, but it might just be that I need to pool my factors
My issue stems from the fact that, for example, for week 3 in 2015 for species A at location X, I have 150 measurements for fruit perimeter, but 75 for fruit mass, only 3 for Total Phenolics, and 25 for Total Soluble Solids. This is due to the different processing of the samples (75 fruit harvested and massed, photographed on two sides to measure perimeter, a subset removed for soluble solids processing, and the rest pooled and used to measure total phenolics in triplicate). Repeat for 5 species at 4 locations during a 9-week harvest period over 2 years. I'm trying to compare these species to determine what differences (if any) exist between species, between the same species at different locations, between years, and changes over time within the harvest season. I haven't worked with such a large and disparate dataset before. Can I just leave blank rows in Excel and it will work out right? I've been having issues with R on just a subset of the data that has consistent sample numbers, but it might just be that I need to pool my factors
I am not an expert in your area but I am intrigued by the data structure. Could you design a basic unit of analysis (say, month, species, location, ... and then attributes).
It looks like you have various different raw rates for assessing the attributes. Why not collect the summary (average, standard deviation) for the basic units of analysis for the attributes. That is, pre-process the raw data to get a consistent no-missing-variables summary of the data. From there I could for example see ANOVA, or log linear model, or really a wide variety of multivariate techniques. I think the problem is that you want to do the analysis on disparate raw data and they are clearly not structured to support that.
Based on what you've said, it sounds like you just have missing data. Organize the individual observations like usual and for whatever measures you don't have, put NA. Then for each measure you can do an analysis that excludes the observations which are missing data for that measure.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Applied Statistics