There has been substantial discussion on RG about the relative merits of the RG score, which tends to favour Reads, Recommendations and Q&A, and Total Research Interest (TRI), which tends to favour Citations, Reads and Recommendations.
The fundamental difference between the RG score and TRI is based on Q&A and Citations.
There is substantial variation in each metric, but casual empiricism would suggest that the RG score and TRI are negatively correlated, as academics tend to specialize in the components that lead to a high RG score or high TRI, but typically not both.
Is it time to use a simple new RG mean score based on the Geometric Mean (GM) of the RG score and TRI:
RG mean score = GM (RG score, TRI) = square root (RG score x TRI).
This sounds like an excellent idea.
There might be a debate about which of the three Pythagorean measures of central tendency, namely the arithmetic, geometric or harmonic means, might be the most appropriate measure.
As the range of variation of the RG score and TRI can be very different, either the geometric mean or harmonic mean would seem to be appropriate.
To Chia-Lin Chang:
Many thanks for your thoughtful and insightful response regarding the three Pythagorean measures of central tendency.
I am sure that the empirical analysis will support the supposition that the rankings according to the proposed new measure will be entirely different from those based on the RG score and TRI.
Dear Michael John McAleer and Chia-Lin,
Your idea is very challenging and interesting. However, in my opinion,
your proposed formula (involving the geometric mean) for a new the so-called RG mean score
RG mean score = GM (RG score, TRI) = square root (RG score x TRI)
probably would be inappropriate and unpopular for most RG members. Moreover, generally, by this formula, RG mean score of a particular RG member would be "vey small" if at least of one components RG score or TRI is "very small".
However, motivated by a suggestion of Chia-Lin (15 days ago), in order that a new RG mean score to be “compatible” with both recent RG score and recent TRI, it would be suitable to define it as a some other weighted mean in two variables (such as arithmetic mean or harmonic mean) with some weights that would “balance” the variation of the RG score and TRI.
Best regards
To Romeo:
Many thanks for your detailed and clear response to the suggestion of a simple new RG mean score.
It is well known that the arithmetic mean is affected by extreme values, whereas the harmonic mean is far less affected by extremes.
The geometric mean lies in between the arithmetic and harmonic means.
As of 9 May 2019, the highest RG scores and TRI I have seen are:
Highest RG score:
RG score = 954.54; TRI = 1,974; h-index = 27.
It does not matter which of the three means is used as the quantitative answers will be reasonably similar.
Highest TRI:
TRI = 122,213; RG score = 56.36; h-index = 260.
TRI swamps the RG score, and the calculated arithmetic, geometric and harmonic means will be very different, with the RG score effectively being eliminated.
The second case is extreme, but there are many other examples from the medical sciences where TRI > 50,000, with the corresponding RG scores < 55.
Consequently, any averaging of the RG score and TRI would be quite unpopular for academics who are keen on the RG score, for obvious reasons.
Dear Michael John and Chia-Lin,
I completely agree with your ( Michael John's) assertion that “any averaging of the RG score and TRI would be quite unpopular for academics who are keen on the RG score, for obvious reasons.” In addition, this is complicated by the fact that TRI is a is a recent “experiment” on RG. A short interesting observation on TRI was given by the respectable Serbian mathematician Miroslav Pavlović (my follower at RG) in his first answer available at https://www.researchgate.net/profile/Miroslav_Pavlovic/answers
Best regards
Dear Michael John McAleer and Chia-Lin,
Means are my love from my student’s and scholars contestants days. Besides the standard four means, there are some other means (see my unpublished paper (with Miomir Andjić) available at https://arxiv.org/pdf/1804.00542.pdf, also available at https://www.researchgate.net/publication/324167291_Two_curious_inequalities_involving_different_means_of_two_arguments). Also see my paper (with David Kalaj) available at https://www.researchgate.net/publication/50283324_A_Converse_of_Minkowski's_Type_Inequalities, my paper available at https://www.researchgate.net/publication/329629451_Minkowski-Clarkson's_like_inequalities and my preprint (submitted paper) available at https://www.researchgate.net/publication/329701198_A_Geometric_Proof_of_the_Inequality_AM_GMQM_HM_in_Two_Variables.
Best regards
To Romeo:
Many thanks for the illuminating references and updates on the RG score and the experimental TRI.
Regarding the comment on measuring intellectual capability by Miroslav Pavlović, it is like using IQ, which is problematic.
TRI is intended to evaluate academic research quality based on an arbitrary weighting of Citations, Recommendations, and Full Text and Other Reads (that is, not Reads of Q&A).
The arbitrariness of the weights might be problematic, but at least the algorithm is given.
A serious concern is that the h-index is not (presently) used in calculating TRI.
It might be worth noting the following serious issue relating to the RG score:
https://www.researchgate.net/post/A_fatal_flaw_in_the_RG_score_if_you_care
The following thread might be of interest as it is related to this Question:
https://www.researchgate.net/post/Q_A_on_the_RG_score
To Noori Abdul-nabi Nasir :
Many thanks for your encouraging response.
Further to the previous instructive discussions regarding a proposed new RG mean score:
It would seem that any combination of the RG score and TRI would be dominated by TRI, for which the highest observed value is greater than 100,000, whereas the highest observed RG score seems to be less that 1,000.
High TRI values are observed frequently in the medical and bio-medical sciences, among others, where TRI in the high 5 figures is common.
TRI is dominated by Citations, so any combination of the RG score and TRI would basically reflect the number of Citations.
It might be worth considering a weighted RG mean score, such as:
RG mean score = GM (RG score, a*TRI) = square root (RG score*a*TRI)
where might be 1% to reduce the extreme;y high TRI.
The Geometric Mean (GM) could be replaced by the Harmonic Mean (HM) in the above formula to reduce the extreme values associated with TRI.
To Chia-Lin Chang:
Many thanks for your insightful comments and suggestions on a new RG mean score.
The following Question should be added to the list, especially as it might well have been the first to invite discussion about the meaning of the RG score:
2013
https://www.researchgate.net/post/Which_researcher_has_the_highest_RG_score_and_what_does_that_really_mean
The following thread might be of interest as it is related to this Question:
https://www.researchgate.net/post/Q_A_on_the_RG_score
Many academics in medical-related research, where team research is standard, have TRI >> 50,000, with the highest I have seen being 122,521, with Citations = 243,462 and RG score = 56.37.
As academic researchers in medical-related disciplines tend not to engage in Q&A, their RG scores seldom exceed 60.
It is worth reflecting on why so many academics seem to emphasize the RG score when the most prolific researchers have relatively low RG scores.
I enjoy the endless discussions in Q&A, where everyone seems to have an opinion.
This should be encouraged as most Questions are open ended and do not have uniquely correct Answers.
In connection with the RG score, as of 26 May 2019 I have compiled a list of 126 academics with RG score > 100.
I do not pretend to have a comprehensive list, but I am able to provide some data to support sensible empirical discussion about the RG score, whether you care about the RG score or not.
In my list, the highest RG score = 965.45, with h-index = 27.
Of the 126 academics, 32 have h-indexes in the range [0, 10).
Two academics have h-index = 0, and four have h-index = 1.
The remaining 94 academics have an h-index in the range [10, 91], with more than one-half of 94 having h-index > 25.
This is not achieved through endless Q&A, contrary to what has been repeated in various threads relating to the RG score, but through serious academic research over an extended period.
For years scientists have complained about how their reputation is measured. Some methods simply don’t reflect and do justice to individual achievements. A good example is the JIF. This is the ratio between the citations of a journal and the sum of articles published in this journal in the previous two years. The JIF is issued by Thomson Reuters yearly, and is meant to serve as an indicator for the prestige of a journal ( https://www.educationdive.com/news/guest-post-researchgates-rg-score-and-what-it-could-mean-to-science/83287/ )
To Noori Abdul-nabi Nasir :
Many thanks for the reference to the introduction of the RG score in 2012.
To Olutosin Ademola Otekunrin :
Thank you for your interest in the interesting academic exchanges regarding this Question.
The following thread might be useful to anyone (still) asking about the RG score: https://www.researchgate.net/post/A_Childs_Guide_to_Calculating_the_RG_Score
This interesting related thread makes it clear that, in the absence of collusion, there is one thing that an academic cannot control, namely all the other academics who might interact with our published articles, unpublished research, projects, Reads, Recommendations, and Q&A.
Therefore, for those who might be interested in increasing their RG scores, for reasons that escape me, there is no guarantee that the RG score will change in the desired direction.
There have been many detailed and carefully considered Answers to this Question regarding the RG score, which is related to a number of interesting threads regarding the calculation, meaning, interpretation, and usefulness, among others, of the RG score.
The RG score is a computer-generated empirical weighted sum of all activities and interactions on RG, excepting Citations and the h-index.
It seems that the RG score depends heavily on Reads and Recommendations which, in turn, is heavily affected by Q&A.
I would like to thank all the academics who have provided interesting and stimulating Answers to the Question.
It might be worth explaining the items that affect the calculation of the RG score.
The presentation in RG is not entirely clear on the definitions of "published articles" and "unpublished research".
The description of the RG score is given at:
https://explore.researchgate.net/display/support/RG+Score
"The RG Score measures scientific reputation based on how your work is received by your peers. We believe that researchers are the best judges of each other's work and that all a [sic] person's research, published or not, deserves credit."
The term "work" is not defined.
A "reasonable" definition of "work" would include, as you have suggested, books, edited monographs, and fully refereed and edited conference proceedings volumes, which would appear under "published articles".
Unpublished versions of books, edited monographs, and conference proceedings volumes would presumably appear under "unpublished research".
I wish to thank all the respondents for their interesting, challenging, informative, instructive, innovative, topical, timely, entertaining, and occasionally highly amusing, Answers.
It might be worth highlighting that an academic's RG score can change even if they do absolutely nothing for an entire week in terms of adding zero Q&A, Reads, Recommendations, published articles, and unpublished papers.
This is because the RG scores of all the other academics who interact with an academic's previous Q&A, Reads, Recommendations, published articles, and unpublished papers, during the week will affect your RG score.
The RG score is an interaction between an academic and every other academic who engages in exchange and interactions with all current and previous RG activities.
The equation to calculate TRI is:
TRI = (0.5 x Citations) + (0.25 x Recommendations) + (0.15 x Full Text Reads)
+ (0.05 x Other Reads)
Reads are by RG members.
The h-index is not used in calculating TRI, so the sign and size of the correlation coefficient between the new RG mean score and the h-index is entirely empirical.
I would like to sincerely thank the two distinguished Professors of Economics @Michael John McAleer and Chia-Lin Chang for giving these informative and helpful tips regarding this discussion. You most appreciated Sir and Ma.
To Olutosin Ademola Otekunrin :
Many thanks for your positive and supportive Answer.
To Chia-Lin Chang:
Many thanks for the explicit equation to calculate TRI.
It might be worth noting that the equation to calculate TRI uses Publications Recommendations.
This differs from the the algorithm in calculating the RG score, which includes Recommendations for Q&A.
The algebraic equation given previously is for calculating TRI, which holds on a weekly basis as well as in aggregate for all previous weeks.
It is a formula that is used to calculate TRI for every academic on RG:
https://www.researchgate.net/application.researchInterest.ResearchInterestHelp.html
As some academics are not calculating TRI correctly, the steps are as follows:
Citations are marked clearly under "Stats overview", with a weight of 0.5.
Recommendations refer only to Publications, with a weight of 0.25, and differ from Recommendations reported under "Stats overview", which also include Recommendations for Q&A.
Reads refer to Full-text Reads (with a weight of 0.15) and Other Reads (with a weight of 0.05) by RG members, and differ from Reads reported under "Stats overview", which also include Reads for Q&A.
Thank you Prof. Chia-Lin Chang for this information on TRI.
- Badges
- Science topic
- Similar topics
- Citations