Dear Anjali Pant ,

Look the link please.

http://www.scielo.br/pdf/brag/v72n4/aop_bragaag0032.pdf

Regards, Shafagat

Modified Mann-Kendall Test For Serially Correlated Data Using Hamed and Rao (1998) Variance Correction Approach

Code Modified Mann-Kendall Test For Serially Correlated Data Usin...

Modified Mann-Kendall Test For Serially Correlated Data Using Yue and Wang (2004) Variance Correction Approach.

Code Modified Mann-Kendall Test For Serially Correlated Data Usin...

To answer your question in Details,

Check Auto Correlation of each series by acf() function in R, or If you are using XLSTAT, you will find the option to plot auto correlation charts in Time-series menu, Descriptive Statistics.

If at a given Confidence interval, the correlation coefficient value is crossing the limit, that series is said to be correlated at that particular lag.

In the attached picture, the series is auto-correlated at lag-2 and lag-4.

Apply mann- Kendall test to only series values that are serially correlated.

One more suggestion, if you happen to have data for more years, consider incorporating more years.

You can use the package 'modifiedmk' from CRAN R repository

Dear Sandeep

Thanks for your response. I understand that if there is auto-correlation in series, then we should apply modified Mann-Kendall. However the problem is how to check the presence of autocorrelation in the series of 600 districts ? And it will be tedious to apply Man-kendall for certain districts and modified mann-kendall for the other set of districts. Hence I wanted to inquire whether blindly modified Man-kendall could be applied to all the 600 districts?

Modified Mann Kendall and Mann Kendall actually produce different results. Hence cannot be applied blindly.

It is easy to do this task in R.

I can work with you on this if you are interested.

An R code can be generated to read all columns of the data sheet, perform Auto-Correlation test and give only outputs that are auto correlated.

I have plans to automate the tasks in my future packages.

In the mean time please send the data to [email protected]

@Lida khoshhal... I am finding it hard to understand your query. Please drop me a detailed message along with sample data.

Hi Dr. Patakamuri

First of all thank you for coming up with "modifiedmk' R-package. However, recent literature by Piyoosh and Ghosh, (2016) and Serinaldi and Kilsby (2016) noted the use of corrected and unbiased trend-free-pre-whitening (TFPWcu) as one of the approaches in addressing aoutocorrelation issue. Is this (TFPWcu) different from your TFPWMK approach?

Thank you so much

Dawit

I have updated the package with Bias-Corrected version of Trend-Free Pre-whitening function. Please Download and explore that. And yes, my TFPWMK is different from the one you have mentioned.

Bias correction suggested by Hamed(2009) is implemented in the recent update if my package. Which is modified version of simple TFPWMK.

Dawit A Shiferaw Consider using Bias corrected TFPW suggested by Hamed (2009) in case if you are using Mann-Kendall based tests. From Serinaldi and Kilsby (2016) paper, you can see that the results of Hamed (2009) found to be very good. (Refer column-2 page4 and figure.1 in page-5.

Please be noted that there is no universal solution to address the issue of Serial Correlation. If you, for some reason insist on using TFPWcu, I may code it and add this functionality later in my package. Keep watching for future updates!

In the mean time, experiment with all the methods available in the package.

Happy trend analysis! :D

Thank you so much! Indeed I tested all the methods in the "modifiedmk' R-package.

Dawit

Can someone help for interpretation of modified Mann kendall test result I got in R please.

I am very confused with the different Zs P-values and variances , please I want to know which ones to use and how to interpret them

Seydouba Soumah One simple way of interpreting the results is, When ever the Z-Value is above or below 1.645 (i.e, plus or minus 1.645), the trend in significant at 90% level. If the Z-Value is above or below 1.96 (i.e, plus or minus 1.96), the trend in significant at 95% level. the Z-Value is above or below 2.57 (i.e, plus or minus 2.57), the trend in significant at 99% level.

@Sandeep Kumar Patakamuri,am interested in the R-code that will automate the acf () across multiple columns,please kindly send to my mail: [email protected]

Many thanks.

Sandeep Kumar Patakamuri am interested in the R-code that will automate the acf() across multiple columns, please kindly send to my mail [email protected]

Many thanks

@Cristian Njoku,

I have not written any code for handling multiple columns at once. I wish I have a readymade script available. Sorry that I can't be of much help here.

Hi Dr. Patakamuri

I would like to ask some advisory in the use of trends with your R code if it´s possible. I would appreciate if you could give me a contact mail to write to you.

Thank you.

Christian Njoku You can use something as follows in R for acf () across multiple columns

data

Anjali Pant Since the problem of auto-correlation doesn't exist, so I think MK is enuf.

In case of a p-value less than 0.05, we need to consider the MMK test.