Tension and Systematics in the Gold06 SnIa Dataset

Abstract

The Gold06 SnIa dataset recently released in astro-ph/0611572 consists of five distinct subsets defined by the group or instrument that discovered and analyzed the corresponding data. These subsets are: the SNLS subset (47 SnIa), the HST subset (30 SnIa), the HZSST subset (41 SnIa), the SCP subset (26 SnIa) and the Low Redshift (LR) subset (38 SnIa). These subsets sum up to the 182 SnIa of the Gold06 dataset. We use Monte-Carlo simulations to study the statistical consistency of each one of the above subsets with the full Gold06 dataset. In particular, we compare the best fit $w(z)$ parameters (w_0,w_1) obtained by subtracting each one of the above subsets from the Gold06 dataset (subset truncation), with the corresponding best fit parameters (w^r_0,w^r_1) obtained by subtracting the same number of randomly selected SnIa from the same redshift range of the Gold06 dataset (random truncation). We find that the probability for (w^r_0,w^r_1)=(w_0,w_1) is large for the Gold06 minus SCP (Gold06-SCP) truncation but is less than 5% for the Gold06-SNLS, Gold06-HZSST and Gold06-HST truncations. This result implies that the Gold06 dataset is not statistically homogeneous. By comparing the values of the best fit (w_0,w_1) for each subset truncation we find that the tension among subsets is such that the SNLS and HST subsets are statistically consistent with each other and `pull&#39; towards LCDM (w_0=-1,w_1=0) while the HZSST subset is statistically distinct and strongly `pulls&#39; towards a varying w(z) crossing the line $w=-1$ from below (w_0<-1,w_1>0). We also isolate six SnIa that are mostly responsible for this behavior of the HZSST subset.