Research into second or foreign language (L2) learning has demonstrated that L2 anxiety, perceived proficiency, and L2 willingness to communicate (L2WTC) profoundly impact language learning outcomes. However, the complex interplay between these variables has yet to be fully explored, as these factors are dynamic and context-specific and can vary across different learners and learning environments. The study, therefore, utilized a cross-sectional quantitative survey research design to scrutinize the causal relationships between L2 anxiety, English proficiency, and L2WTC of 609 Ethiopian preparatory school students. The model for the L2WTC, both inside and outside the classroom, has been expanded to include an additional sub-scale known as the L2WTC in a digital setting. Moreover, in contrast to the commonly recognized debilitative-focused L2 anxiety, the construct of L2 anxiety has been divided into facilitative and debilitative anxiety. This method allows us to measure not only the presence or absence of anxiety but also evaluate if anxiety helps or hinders the L2 learning experience. A self-assessment proficiency measure was also developed specifically for Ethiopian high school students. The study treated facilitative and debilitative anxiety as independent variables while considering English proficiency and L2WTC in the classroom, outside the classroom, and in digital settings as dependent variables. English proficiency was also used as an independent variable to predict L2WTC in these three settings. Our proposed model, including these variables, was tested using structural equation modelling (SEM). According to the descriptive analysis, the mean scores of L2WTC in the three settings were generally low, ranging from 2.30 to 2.84. Debilitative anxiety casts a shadow on the positive aspects of anxiety. The English proficiency was also too low. According to SEM, debilitative anxiety displayed a statistically significant negative impact on L2WTC inside and outside the classroom, in digital settings, and English proficiency. In contrast, facilitative anxiety was found to positively contribute to L2WTC outside the classroom, in digital settings, and in English proficiency. English proficiency made a statistically significant and positive contribution to L2WTC within and outside the classroom and in digital contexts. L2WTC inside the classroom positively contributed to L2WTC outside the classrooms and in digital contexts. We systematically compared the findings with existing studies in the discussion and elucidated the pedagogical implications, limitations, and potential future research avenues. The outcomes of our study have the potential to significantly contribute to the advancement of theoretical and empirical knowledge about improving English education, learning, and communication not only in Ethiopia but also in similar contexts of English as a foreign language (EFL).