Difference between revisions of "Data imputation"
| (One intermediate revision by the same user not shown) | |||
| Line 1: | Line 1: | ||
| − | A variety of '''Imputation techniques''' are available. The ''k''NN estimation method is one, or actually several (depending on the type of information being used, the definition of nearness, the weighting procedure, etc.). Imputation can be regarded as the opposite of amputation; information in a specific format is added to ensure functionality. A straightforward way is to impute plots from a reference data set to a stand described by totals and average values in a register. The imputation is done when values at the stand-level corresponds to those at the plot-level. Unrealistic stands, with more or less no variation (too homogenous), can be estimated if 1. only one plot per stand is imputed, 2. if more than one plot per stand is imputed but the reference data set is "complete" (all possible plots are represented) and a specific plot can be imputed to the same stand more than once, or 3. if no measures are taken into account, to secure a certain variation. | + | A variety of '''Imputation techniques''' are available. The ''k''NN estimation method is one, or actually several (depending on the type of information being used, the definition of nearness, the weighting procedure, etc.). Imputation can be regarded as the opposite of amputation; information in a specific format is added to ensure functionality. A straightforward way is to impute plots from a reference data set to a stand described by totals and average values in a register. The imputation is done when values at the stand-level corresponds to those at the plot-level. Unrealistic stands, with more or less no variation (too homogenous), can though be estimated if 1. only one plot per stand is imputed, 2. if more than one plot per stand is imputed but the reference data set is "complete" (all possible plots are represented) and a specific plot can be imputed to the same stand more than once, or 3. if no measures are taken into account, to secure a certain variation. |
| − | + | Therefore, a desired within stand-variation regarding certain variables can be stated prior to imputation. Another way to obtain more realistic representation of a stand by imputed plots is by the [[Import_of_FMPP_data|"Forking method"]], a part of the FMPP. | |
| + | |||
| + | [Is it the imputation of age and height to calliper-trees, as in FMPP? But isn't this handled within the system, following settings in control tables?] | ||
Latest revision as of 11:07, 9 September 2009
A variety of Imputation techniques are available. The kNN estimation method is one, or actually several (depending on the type of information being used, the definition of nearness, the weighting procedure, etc.). Imputation can be regarded as the opposite of amputation; information in a specific format is added to ensure functionality. A straightforward way is to impute plots from a reference data set to a stand described by totals and average values in a register. The imputation is done when values at the stand-level corresponds to those at the plot-level. Unrealistic stands, with more or less no variation (too homogenous), can though be estimated if 1. only one plot per stand is imputed, 2. if more than one plot per stand is imputed but the reference data set is "complete" (all possible plots are represented) and a specific plot can be imputed to the same stand more than once, or 3. if no measures are taken into account, to secure a certain variation.
Therefore, a desired within stand-variation regarding certain variables can be stated prior to imputation. Another way to obtain more realistic representation of a stand by imputed plots is by the "Forking method", a part of the FMPP.
[Is it the imputation of age and height to calliper-trees, as in FMPP? But isn't this handled within the system, following settings in control tables?]