Dear All,
Thanks for all the replies, based on which we can say that "Systematic grid sampling does belong to probability sampling." There is a minor condition that "the grid is laid down at random" which appears to be a little bit controversial. Generally the grid is laid down based on the latitude-longitude system or a projected national grid system. Since there is no specific relationship between the coordinate system and geochemistry, I would like to say it does satisfy the requirement of randomness. Best regards, Chaosheng -------------------- Dr. Chaosheng Zhang Lecturer in GIS Department of Geography National University of Ireland, Galway IRELAND Tel: +353-91- 49 2375 Fax: +353-91- 49 5505 Email: [EMAIL PROTECTED] Web: http://www.nuigalway.ie/geography/gis ----------------------------- Original message: --------- Dear all, In geochemical investigations, it seems that there is a growing trend to use a systematic sampling scheme based on a predefined grid system. Here is the question: Does such a systematic grid sampling belong to probability sampling or not? I have got a comment on this topic: "Here, there are no probability features of the geochemical variables because there is no chance experiment involved." Cheers, Chaosheng ------------- Replies: ----------------- From: Yetta Jager [EMAIL PROTECTED] Dear Chaosheng, My understanding is that it is probability sampling if you lay the grid down at random because there is a known probability of selecting each location as part of the sample. The random placement of the first location determines the rest, but the probability can still be calculated. Personally, I don't like using such a design when spatial statistics are involved because you won't get the closer lags. Yetta ------ From: Max Kummerow [EMAIL PROTECTED] It's better than chance (random sampling) because certain improbable unrepresentative samples are excluded from the possible sample set. As a result the grid estimates are more representative than random samples. I've forgotten the reference where I read about this but googling "systematic sampling" will get you started on finding information about calculating the increase in efficiency in a systematic sample. An exception to the above would be if there is a pattern in the data that is picked up by the grid sample, but that seems unlikely in geochemical systems. ------- From: Jan W Merks [EMAIL PROTECTED] Hello Chaosheng, Systematic grid sampling does indeed meet the requirement for true probability sampling. Rgds, JanWM ------ From: Mat Pawley [EMAIL PROTECTED] Hi Chaosheng, It depends - if the systematic sample (SYS) grid is randomly positioned, then all (N) points in your sample extent have equal probability of selection (i.e. n/N), and therefore your estimate of the population mean is unbiased. However, all pairs of samples do not have equal probability of selection (for instance, neighbouring points in the sample extent cannot ever be selected), leading to difficulty in the estimation of the standard error (this, of course, is under the assumption that the population is [positively] spatially autocorrelated - if the population was independent, then the position of the samples does not really matter - i.e. a systematic sample is effectively the same as SRS). The difficulty in determining inference for SYS using design-based methods has meant that model-based methods (e.g. geostatistics), where randomness arises from the stochastic spatial process described by the model, are sometimes used. Mind you, in ecology, the most common-method of treating SYS inference still tends to be to pretend that the data came from SRS .. this gives a conservative estimate, i.e. over-estimate, of the standard error). Regards, Mat --- Dr Matthew Pawley Post-Doctoral Fellow Leigh Marine Lab University of Auckland ---------------- From: [EMAIL PROTECTED] Dear Chaosheng You said here is the equation but did not provide the equation itself. As for optimum grid sampling, we are reaching a conclusion that the question itself is problematic, let alone the answer. believe it or not, almost all measurement in one way or another is contaminated by scale issue either in time or space. You may want to look at our paper in journal of hydrology vol. 318 entitled "On depressional storages: The effect of DEM spatial resolution" Hope this helps. Thanks Abedini
