Understanding speech recognition errors, especially those related to accents, is challenging due to the complexity of the models and scarcity of data. This paper addresses this issue by exploring the use of synthetic data to investigate accent-related variations and their impact on Automatic Speech Recognition (ASR) performance. We synthesise Spanish-accented English and compare the speech features captured by synthetic speech with those found in natural speech. We generate speech with phoneme-level variation using Spanish voice synthesis and phoneme-to-speech synthesis and then assess ASR sensitivity to such variations. Our findings show that synthetic data captures phonemic patterns of Spanish well, suggesting its utility, coupled with ASR, in L1-L2 phonemic difference modelling. In contrast, phonotactic patterns are not captured to the same extent by synthetic data. We also show that the variants built from the synthetic data accurately challenge ASR systems, prompting a potential method for testing and enhancing ASR accent robustness and explainability for speech research.