Naslov
Comparison of Pearson's and Spearman's correlation coefcients for grain cadmium concentration and agronomic traits in winter wheat genotypes

Autori
Rebekić, Andrijana ; Lončarić, Zdenko ; Petrović, Sonja ; Marić, Sonja

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, sažetak, znanstveni

Izvornik
Book of abstracts BIOSTAT 2014 - 21st International Scientifc Symposium on Biometrics, / Jazbec, Anamarija ; Pecina, Marija ; Sonicki, Zdenko ; Šimić, Diana, Vedriš, Mislav - Zagreb : Croatian Biometric Society c/o University of Zagreb, School of Medicine, School of Public Health ”Andrija Štampar”, 2014, 25-25

Skup
BIOSTAT 2014 - 21st International Scientific Symposium on Biometrics

Mjesto i datum
Dubrovnik, Hrvatska, 11.06.2014. - 14.06.2014

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
correlation ; wheat grain ; agronomic trais

Sažetak
Winter wheat (Triticum aestivum L.) is staple crop in human nutrition and therefore lots of resources is invested in wheat research resulting in tens of thousands publications covering results obtained from wheat research annualy. In many publications Pearson’s correlation coefficient is reported for variables that are, by its nature or by high number of outliers, non-normally distributed. The aim of this paper is to compare Pearson’s productmoment correlation coefficient (r) and Spearman’s rank correlation coefficient (rs) in examination of relation between grain cadmium (Cd) concentration (mg kg-1) and grain Zn (Zn) concentration (mg kg-1), plant height (cm), plant weight (g), number of spikelets per spike and 1000 kernel weight (g) in 51 winter wheat genotype. Experiment was set up in semi- controlled conditions, 40 wheat grains was sown in plastic containers containing 12 kg of Cd contaminated soil (20 mg Cd kg-1 soil) or uncontaminated soil (0 mg Cd kg-1 soil). Experimental design was completely randomized design with four replicates. Statistical analyses were done using SAS 9.3. Software for Windows, Copyright © 2012 by SAS Institute Inc., Cary, NC, USA, SAS Enterprise Guide 5.1 and SAS JMP® 9.0.2. Distribution of variables was tested using Shapiro-Wilk test and non- normally distributed variables were log transformed to assess normality. Homoscedasticity was checked with scatterplot of the residuals versus predicted values. Original data (prior to removal of outliers and log transformation) were correlated and r and rs coefficients were compared. Pearson’s correlation coefficient is a measure of strength of linear relationship, so higher r coefficient in compare to rs indicating linear rather that monotonic relationship. On uncontaminated soil, correlation analysis of original data yielded similar r and rs coefficients for correlation between grain Cd and grain Zn, plant weight and number of spikelets per spike. These three dependent variables were normally distributed according to Shapiro-Wilk test of normality, and that is a probable reason for similar values of these two coefficients. On uncontaminated soil, variable plant height was non-normally distributed (p = 0.002) even after log transformation (p = 0.021). The r coefficient for grain Cd concentration versus plant height, was non-significant (r = -0.22 ; p = 0.108) while Spearman’s correlation coefficient was marginally significant (rs = -0.28 ; p = 0.04) although both correlation coefficients indicated very poor correlation. After log transformation and outlier removal (only one outlier) correlation coefficients increased, r = -0.31 ; p = 0.030 and rs = -0.32 ; p = 0.023, and both were significant at significance level of 95%. Correlation between grain Zn and Cd concentration on Cd contaminated soil was marginally significant according to rs while r obtained non-significant correlation between these variables. After log transformation of grain Zn concentration, correlation coefficients remain very low, with rs still obtaining marginally significant relation. This result indicating that relation between grain Cd and Zn concentration is probably monotonic rather than linear. Choosing appropriate measure of correlations is crucial for assessing true nature of relationship between variables. Normally distributed, homoscedastic variables, measured on interval scale are appropriate for Pearson correlation, while ordinal, non-normally distributed variables will perform better with Spearman’s correlation coefficient.

Izvorni jezik
Engleski

Znanstvena područja
Poljoprivreda (agronomija)

POVEZANOST RADA