6. Non sampling errors

Transcript

Welcome to the last lecture of week two. Here we are going to talk about Non-Sampling Errors. Non-Sampling Errors are errors after you have acquired the data, now you are processing the data and what are all the errors which can get into it and how is that going to influence the output. So, these things are called as non-sampling errors. Non-Sampling Errors as the name suggest that those errors that come in due to external reasons. It refers to any deviation between survey result, and truth which are not a result of random selection of observation. So non sampling error which cause external focus and here you will try to get the solution study mechanism design. So, in non-sampling error you will have random error and you will have systematic error. Random error are suppose, you have yes you have a plot y and x so you are trying to draw a line. So, the errors are it goes like, this a data which does not fall on the straight line but slightly deviated from the straight line and the distance need not be uniform so, this is a random error, and if what is systematic error if you have a line like this some interpretation of a result like this and if the result is offsetted in this way or this way then there is a systematic shift of the data, and the error can be easily identified and removed just by minusing it or adding it some data to the original data to get the output so, that is called as Systemic error so Random error this is random and this is systemic error. So, this is called as systematic error.

So, let us understand the Coverage Error, errors due to omission that means to say you are under covering or enormous, inclusion duplication, misclassification, of units in the survey frame leads to this coverage error. Either you take very little data in the frame, or you take more data in the frame, try to measure the same sample and add more data, or miss classifications all these things lead up to coverage error, affects every estimate produced by the survey thus most important type of error is coverage error, you should be very very careful about it. You cannot measure the same sample and write it 10 times so that is duplication. You cannot write out of thousand trees i picked three trees not accepted, out of thousand trees i try to get 600 trees not accepted, out of thousand trees of apple I want to have but I have here only grapes orchard so can I, no, its misclassification. So, all these things are there. It can have both Spatial and Temporal dimensions and cause bias. It can have both spatial and temporal. Usually, due to under coverage this coverage error always happens.

The Non Response Errors – the error that occurs when the experiments or survey fails to get a response to one or possibly all of the questions is called as non-response error. Non response error also is quite common when you do the experiments unknowingly, when you start collecting the data unknowingly, when you do not do pilot experiments, you do pilot experiments improve your questionnaire. So, when you do that this non-response error will be reduced. Doing experiments on your field is not so easy you have to understand the concept and then do. The error that occurs when the experiments or survey fails to get a response to one or possibly all of the questions is called as non-response error. This non response error again it can be classified into two – one is called as total non-response error, the other one is called as partial non response or hundred percent or fifty percent so, that is called as the non response error classification. There can be processing error, processing error or error when pops up when you are trying to get the data processed in a software or processed into a table form or a graph form to understand the result. Why do you understand because you have to do the reporting of the result. So processing error, it includes all data processing activity after collection and prior to estimation. Such as errors in data capture coding. Many a times what happens we try to give a coding. Why do we do coding, because if you say apple trees means, every time if you say apple trees immediately you will say ok, apple trees means at this point of time will not give you a good deal. So, you will try to be biased. In order to remove it we try to give coding. Then the coding has to be uncoded and then sometimes we do editing, and then we try to do tabulation. If in the tabulation it is pretty interesting, or in the graph also it is pretty interesting. Suppose, let us assume you have data from 0 to 10 and then if you have data from 1000 to 10000. If you have data, if you wanted to put it in linear scale, and if you want to do it with the unit scale multiplication, you will not be able to represent the data. So, the same way tabulation of data is also very very important. So, tabulation of data as well as in the assignment of survey weightage all these things lead to processing error.

Processing error can be of major three categories one is called as coding error, the other one is called as data capturing error, and the third one is going to be editing and imputational error. So, these are some of the errors which get into processing error. So, apart from these errors there are several other non processing errors due to shortage of time i will not be able to go through all these things. However i will try to give a very very brief explanation on all these things. Failure to locate some individuals. See after you do the measurement and then you put all these things into a small box and then keep the sample for further analysis or if you are not able to identify the sample which you measured and recorded so, that is because of failure to look at some individuals, and when you are trying to do all this farm experiments, you have to mark the tree or the plant where in which you are trying to do the study, and this has to be repetitively measured at regular intervals of time. You have to measure the same tree. So, failure to locate the some individuals, error of measurements on units for example grams, milligrams, then you have centimeter, millimeter, all those things non response bias. So, this is something which is very typical to handle then Careless data collection, Error of classification, Data entry error, Inappropriate analysis, Misrepresentation of the facts and Telescopic, referencing period. So, all these things are other non-response errors.

In this particular lecture we covered several of the topics on Non sampling errors. We first saw random error, systematic error. Then we went into Coverage error, we went into Non response error then, processing error, and finally a big list of non-response errors have been discussed.

Thank you very much.