I’m just performing an analysis over some supplier’s bids, and my objective is to spot any unsual relation between bidders. In this case, I have done some graphics in powerbi which are visually effective, but my intention is to capture these cases through a knime workflow. In the picture presented below there are bid rounds (X axis) and bid amounts (y axis). Each line is a vendor.
In this example , there is a clear relation in the way bids are changing among suppliers (they almost all increase or decrease in the same rounds, and in similar amounts) I have been trying in KNIME the “Linear Correlation” node, but it gives me some contradictory results (probably I’m not configuring it properly). My input data table is very simple:
Round
Vendor 1
Vendor 2
Bidd amount 1
Bidd amount 2
0
AAAA
BBBB
729.450
767.550
1
AAAA
BBBB
729450.0
1.022.400
2
AAAA
BBBB
915.342
1.123.284
3
AAAA
BBBB
925.272
1.342.899
…
…
…
…
…
0
CCCC
DDDD
…
…
1
CCCC
DDDD
…
…
2
CCCC
DDDD
…
…
Which is the best way to measure this kind of relation?
Hello @Sgomezba
Looking at your chart, you are trying to compare cases that are different ‘animal’ or ‘size’, but you want to get insights from vendors behavior.
My first thought -aiming to simplify- would be to normalize the data for each bid or set (animal) -if size, it could be normalize by unit first (?)-, assuming normal distribution (z-score is available in Normalizer node). Then you can try linear correlation within normalized data. If your sample data is as short as 3 to few, student’s t would be more appropriate approach, but as far as I know there isn’t a node or component (?) doing the job; so coding would be a must.
But reality becomes sometimes much more complicated; it would happen that your distribution isn’t normal, then you would want to analyze the trends and apply some other distribution.
Based on these elements, you could go further and test hypothesis for each vendor’s bias…
Interesting challenge by the way. But here, context matters; then I think that the provided sample is too short for the task
