help
help > RE: second-level analysis in Conn
Oct 1, 2015 05:10 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
Hi Alfonso,
Thank you very much for all your help. I have three questions and deeply appreciate your help in confirming these issues before we submit the results for publication purposes. Just to refresh your mind, we examined the correlation between functional connectivity and core symptoms of autism spectrum disorder in three age-groups of participants. This is how I analyzed the data - followed by my three questions:
I selected 6 regressors (three age-groups and their corresponding scores on a given clinical measure) from the "subject effects" list in the 2nd-level results (ROI to ROI), entered the contrast [0 0 0 1 0 0;0 0 0 0 1 0; 0 0 0 0 0 1] in the "between-groups contrasts", then selected a single ROI from our list of ROIs and ran the contrast. I repeated this process for each ROI in our list (for each symptom - total of 5 symptoms).
1) Are the results shown in the "analysis results" table corrected for multiple comparisons, given the route I took explained above?
2) Given that the symptoms severity scores are not standardized, meaning a score of "0" indicates the absence of that symptom, not the average, is our interpretation (e.g. functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour in our third group) correct in the attached graph?
3) When we see a significant effect in one of our three age-groups, how can I run pairwise comparisons (following an interaction effect) to report the p-value associated with that particular effect? For example, in the attached graph, there's significant correlation between connectivity and repetitive behaviour only in our third group, how can I determine the p-value for this particular effect?
Many thanks again,
Diana.
Originally posted by Diana Parvinchi:
Thank you very much for all your help. I have three questions and deeply appreciate your help in confirming these issues before we submit the results for publication purposes. Just to refresh your mind, we examined the correlation between functional connectivity and core symptoms of autism spectrum disorder in three age-groups of participants. This is how I analyzed the data - followed by my three questions:
I selected 6 regressors (three age-groups and their corresponding scores on a given clinical measure) from the "subject effects" list in the 2nd-level results (ROI to ROI), entered the contrast [0 0 0 1 0 0;0 0 0 0 1 0; 0 0 0 0 0 1] in the "between-groups contrasts", then selected a single ROI from our list of ROIs and ran the contrast. I repeated this process for each ROI in our list (for each symptom - total of 5 symptoms).
1) Are the results shown in the "analysis results" table corrected for multiple comparisons, given the route I took explained above?
2) Given that the symptoms severity scores are not standardized, meaning a score of "0" indicates the absence of that symptom, not the average, is our interpretation (e.g. functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour in our third group) correct in the attached graph?
3) When we see a significant effect in one of our three age-groups, how can I run pairwise comparisons (following an interaction effect) to report the p-value associated with that particular effect? For example, in the attached graph, there's significant correlation between connectivity and repetitive behaviour only in our third group, how can I determine the p-value for this particular effect?
Many thanks again,
Diana.
Originally posted by Diana Parvinchi:
Hi Alfonso,
Thank you very much for your reply, as always, we found it very helpful. To answer your question, a repetitive score of "0" here would mean an absence of this symptom - the higher the score the greater the symptom severity. Given this point, I think you're confirming that we are correct in interpreting this finding as showing a negative correlation - as connectivity between these two ROI's increases, repetitive symptom severity decreases. My question is if we want to report this finding (and other similar ones), where can I get the statistical information? Should we take the information from the "Analysis results"? In other words, what is equivelent to running pair-wise comparisons to identify where a significant effect is coming from? In this case, we can see that this effect is significant only in the third age-group from the graph, but what is the p-value corresponding to it? Thanks again for your help.
Cheers,
Diana.
Originally posted by Alfonso Nieto-Castanon:
Thank you very much for your reply, as always, we found it very helpful. To answer your question, a repetitive score of "0" here would mean an absence of this symptom - the higher the score the greater the symptom severity. Given this point, I think you're confirming that we are correct in interpreting this finding as showing a negative correlation - as connectivity between these two ROI's increases, repetitive symptom severity decreases. My question is if we want to report this finding (and other similar ones), where can I get the statistical information? Should we take the information from the "Analysis results"? In other words, what is equivelent to running pair-wise comparisons to identify where a significant effect is coming from? In this case, we can see that this effect is significant only in the third age-group from the graph, but what is the p-value corresponding to it? Thanks again for your help.
Cheers,
Diana.
Originally posted by Alfonso Nieto-Castanon:
Hi
Diana,
Yes, your interpretation is perfectly correct. Connectivity between right Intracalcarine and right FO cortex appears negatively correlated with symptom scores (i.e. higher connectivity between these regions in children with lower symptom scores). In addition there also appear to be an interaction with age, with stronger associations with symptom scores (and perhaps also higher average connectivity as well, see point below) in older children (age group 3) compared to younger children.
There is only one subtlety here that affects the interpretation of the first three bars. Could you please let me know whether the symptom-score covariates are centered or not? If they are centered, did you center them individually within each age-group or globally across all subjects?; and if they are not centered, is a symptom-score equal to zero meaningul/interpretable? This affects the interpretation of the sign and effect-size of the first three bars in your plot -in other words, whether on average the connectivity between right ICC and right FO is positive or negative/anticorrelated-, which in turn affects the correct interpretation of the associations with symptom scores -e.g. higher connectivity in children with lower symptoms may reflect stronger/more-positive connectivity in these children or weaker/less-negative conectivity in these children-. Generally the effect-sizes in your first three bars represent the average level of connectivity within each age-group at the zero level of your symptom score covariate. If the symptom score covariates are centered globally (subtracted the average across all of your subjects) then that zero-level represents the average symptom-level across all of your subjects (and the same level across the three age-groups), if they are centered separately (subtracted the average within each age-group separately) then that zero-level represents the average symptom-level within each age-group separately, and if they are not centered (they represent "raw" symptom scores) then that zero-level represents an actual zero-value of your original symptom variables (again the same level across the three age-groups).
Thanks
Alfonso
Originally posted by Diana Parvinchi:
Yes, your interpretation is perfectly correct. Connectivity between right Intracalcarine and right FO cortex appears negatively correlated with symptom scores (i.e. higher connectivity between these regions in children with lower symptom scores). In addition there also appear to be an interaction with age, with stronger associations with symptom scores (and perhaps also higher average connectivity as well, see point below) in older children (age group 3) compared to younger children.
There is only one subtlety here that affects the interpretation of the first three bars. Could you please let me know whether the symptom-score covariates are centered or not? If they are centered, did you center them individually within each age-group or globally across all subjects?; and if they are not centered, is a symptom-score equal to zero meaningul/interpretable? This affects the interpretation of the sign and effect-size of the first three bars in your plot -in other words, whether on average the connectivity between right ICC and right FO is positive or negative/anticorrelated-, which in turn affects the correct interpretation of the associations with symptom scores -e.g. higher connectivity in children with lower symptoms may reflect stronger/more-positive connectivity in these children or weaker/less-negative conectivity in these children-. Generally the effect-sizes in your first three bars represent the average level of connectivity within each age-group at the zero level of your symptom score covariate. If the symptom score covariates are centered globally (subtracted the average across all of your subjects) then that zero-level represents the average symptom-level across all of your subjects (and the same level across the three age-groups), if they are centered separately (subtracted the average within each age-group separately) then that zero-level represents the average symptom-level within each age-group separately, and if they are not centered (they represent "raw" symptom scores) then that zero-level represents an actual zero-value of your original symptom variables (again the same level across the three age-groups).
Thanks
Alfonso
Originally posted by Diana Parvinchi:
Hi Alfonso,
Could I please get your input on a graph that I have attached. Just to remind you of our analysis, we are looking at the correlation between functional connectivity and symptoms severity in children with autism spectrum disorder (ASD). We have three age-groups of cohorts with ASD. I have attached a screen shot of a graph showing 6 regressors. The first 3 are the age-groups and the last three are the symptom scores within each group. My interoperation of this graph is that functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour. Meaning, as connectivity between these two regions increases, repetitive behaviour decreases. Is this correct? Also, if we'd like to report this finding in a manuscript, how could we provide details of this effect (p-value, beta which are specific to this significant effect)? Many many thanks for your help!
Best,
Diana.
Could I please get your input on a graph that I have attached. Just to remind you of our analysis, we are looking at the correlation between functional connectivity and symptoms severity in children with autism spectrum disorder (ASD). We have three age-groups of cohorts with ASD. I have attached a screen shot of a graph showing 6 regressors. The first 3 are the age-groups and the last three are the symptom scores within each group. My interoperation of this graph is that functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour. Meaning, as connectivity between these two regions increases, repetitive behaviour decreases. Is this correct? Also, if we'd like to report this finding in a manuscript, how could we provide details of this effect (p-value, beta which are specific to this significant effect)? Many many thanks for your help!
Best,
Diana.