help > Strong connections at high thresholds
Showing 1-7 of 7 posts
Apr 28, 2018 10:04 PM | hex_dex - AUTH
Strong connections at high thresholds
Dear Andrew,
You have been very helpful in another topic I had started but I realised that I have wrongly processed my raw data, so I started frpm the beginning. My problem now is the opposite than the last time, I get really strong connections. To be specific, I have 2 groups of interest and 2 time points. I perform a Time x Group analysis using F-test, a=0.001, I test 22 nodes and even at a very high threshold (eg 10) I get really strong connections (eg 18 nodes, 38 edges as in the picture) with p-values equal to zero. I attach a picture of a graph for a specific rhythm tested (a rhythm) and a picture of the fields I fulfill and the results of the p-values. Do you think this is normal?
Additionally when I perform FDR (50000 permutations) I get for the a rhythm results quite similar to that of a threshold=10. Does this mean that everything is ok? I attach the graph for FDR,too.
Last but not least, in order to perform a main effect of time analysis do I need to perform a t-test or am I ok with the f-test?
Thank you,
Katerina
You have been very helpful in another topic I had started but I realised that I have wrongly processed my raw data, so I started frpm the beginning. My problem now is the opposite than the last time, I get really strong connections. To be specific, I have 2 groups of interest and 2 time points. I perform a Time x Group analysis using F-test, a=0.001, I test 22 nodes and even at a very high threshold (eg 10) I get really strong connections (eg 18 nodes, 38 edges as in the picture) with p-values equal to zero. I attach a picture of a graph for a specific rhythm tested (a rhythm) and a picture of the fields I fulfill and the results of the p-values. Do you think this is normal?
Additionally when I perform FDR (50000 permutations) I get for the a rhythm results quite similar to that of a threshold=10. Does this mean that everything is ok? I attach the graph for FDR,too.
Last but not least, in order to perform a main effect of time analysis do I need to perform a t-test or am I ok with the f-test?
Thank you,
Katerina
Apr 28, 2018 10:04 PM | hex_dex - AUTH
RE: Strong connections at high thresholds
I cannot upload 2 or more files in the same post so I upload them
here.
Apr 29, 2018 09:04 AM | Andrew Zalesky
RE: Strong connections at high thresholds
Hi Katerina,
it is difficult to diagnose this potential problem over a forum. I suggest to undertake some basic sanity checks, such as selecting one of the significant connections and checking whether a significant interaction is still evident when manually computing the ANOVA (i.e. with SPSS or online ANOVA calculator) for the connection.
Or generate plots for the data points corresponding to one of the significant connections and show the 2 groups and 2 points with different colors. Is an interaction visually evident in the plot? If no, then it would seem something has gone wrong.
Andrew
Originally posted by hex_dex:
it is difficult to diagnose this potential problem over a forum. I suggest to undertake some basic sanity checks, such as selecting one of the significant connections and checking whether a significant interaction is still evident when manually computing the ANOVA (i.e. with SPSS or online ANOVA calculator) for the connection.
Or generate plots for the data points corresponding to one of the significant connections and show the 2 groups and 2 points with different colors. Is an interaction visually evident in the plot? If no, then it would seem something has gone wrong.
Andrew
Originally posted by hex_dex:
Dear Andrew,
You have been very helpful in another topic I had started but I realised that I have wrongly processed my raw data, so I started frpm the beginning. My problem now is the opposite than the last time, I get really strong connections. To be specific, I have 2 groups of interest and 2 time points. I perform a Time x Group analysis using F-test, a=0.001, I test 22 nodes and even at a very high threshold (eg 10) I get really strong connections (eg 18 nodes, 38 edges as in the picture) with p-values equal to zero. I attach a picture of a graph for a specific rhythm tested (a rhythm) and a picture of the fields I fulfill and the results of the p-values. Do you think this is normal?
Additionally when I perform FDR (50000 permutations) I get for the a rhythm results quite similar to that of a threshold=10. Does this mean that everything is ok? I attach the graph for FDR,too.
Last but not least, in order to perform a main effect of time analysis do I need to perform a t-test or am I ok with the f-test?
Thank you,
Katerina
You have been very helpful in another topic I had started but I realised that I have wrongly processed my raw data, so I started frpm the beginning. My problem now is the opposite than the last time, I get really strong connections. To be specific, I have 2 groups of interest and 2 time points. I perform a Time x Group analysis using F-test, a=0.001, I test 22 nodes and even at a very high threshold (eg 10) I get really strong connections (eg 18 nodes, 38 edges as in the picture) with p-values equal to zero. I attach a picture of a graph for a specific rhythm tested (a rhythm) and a picture of the fields I fulfill and the results of the p-values. Do you think this is normal?
Additionally when I perform FDR (50000 permutations) I get for the a rhythm results quite similar to that of a threshold=10. Does this mean that everything is ok? I attach the graph for FDR,too.
Last but not least, in order to perform a main effect of time analysis do I need to perform a t-test or am I ok with the f-test?
Thank you,
Katerina
Apr 30, 2018 08:04 PM | hex_dex - AUTH
RE: Strong connections at high thresholds
Dear Andrew,
thank you very much for your response. I will perform one of the checks described and see how it goes. Until then, may I ask you something else... Can I run a t-test with 1 in the column corresponding to time by group difference and the rest 0s and detect this way if the null hypothesis exists? I do that and I get significant results when I use 1 but no significant ones with -1. Does this mean that the connectivity is higher in the group 1 than group -1?
Thank you very much,
Katerina
thank you very much for your response. I will perform one of the checks described and see how it goes. Until then, may I ask you something else... Can I run a t-test with 1 in the column corresponding to time by group difference and the rest 0s and detect this way if the null hypothesis exists? I do that and I get significant results when I use 1 but no significant ones with -1. Does this mean that the connectivity is higher in the group 1 than group -1?
Thank you very much,
Katerina
May 3, 2018 06:05 AM | Andrew Zalesky
RE: Strong connections at high thresholds
Hi Katerina,
Yes - this is correct. You can use a t-test to assess the group x time interaction. For the contrast, you should place a 1 or -1 in the interaction column and 0s everywhere else. Depending how you have coded your design matrix, a significant result with 1 in the contrast vector means greater connectivity in group 1 at time 2, whereas -1 means greater connectivity in group 2 at time 2.
Andrew
Originally posted by hex_dex:
Yes - this is correct. You can use a t-test to assess the group x time interaction. For the contrast, you should place a 1 or -1 in the interaction column and 0s everywhere else. Depending how you have coded your design matrix, a significant result with 1 in the contrast vector means greater connectivity in group 1 at time 2, whereas -1 means greater connectivity in group 2 at time 2.
Andrew
Originally posted by hex_dex:
Dear Andrew,
thank you very much for your response. I will perform one of the checks described and see how it goes. Until then, may I ask you something else... Can I run a t-test with 1 in the column corresponding to time by group difference and the rest 0s and detect this way if the null hypothesis exists? I do that and I get significant results when I use 1 but no significant ones with -1. Does this mean that the connectivity is higher in the group 1 than group -1?
Thank you very much,
Katerina
thank you very much for your response. I will perform one of the checks described and see how it goes. Until then, may I ask you something else... Can I run a t-test with 1 in the column corresponding to time by group difference and the rest 0s and detect this way if the null hypothesis exists? I do that and I get significant results when I use 1 but no significant ones with -1. Does this mean that the connectivity is higher in the group 1 than group -1?
Thank you very much,
Katerina
May 3, 2018 09:05 AM | hex_dex - AUTH
RE: Strong connections at high thresholds
Thank you very much!
Katerina
Katerina