In my quantitative studies (Studies 2 and 5), I followed the quality control guidelines of Dörnyei (2007) by aiming for reliability, measurement validity (i.e., construct validity) and research validity (i.e., internal and external validity). The reliability of the questionnaires was checked by calculating the Cronbach’s alpha values for the multi-item scales, while construct validity was sought for by running Principal Component Analysis (PCA) on the scales to check whether the items load onto the same dimension. The internal validity of the questionnaire was ensured by subjecting the first drafts of the questionnaires to peer debriefing and expert judgement, upon which various refinements were made to the instruments. As has been mentioned earlier, in Study 5, both a Hungarian and English version of the questionnaire were created; in order to ensure that the two are equivalents of each other, the English version was examined by a fellow PhD student and was subjected to expert review (Dörnyei & Csizér, 2012).

In the large-scale questionnaire study (Study 2), sampling parameters were established (see Section 3.2.2.3), and participants meeting these parameters were sought for using convenience and snowball sampling. As Dörnyei (2007) formulates, because of the “compromised nature of non-probability sampling” (p. 99), during the discussion of the results it is important to address the issue of external validity, i.e., generalisability. Following Dörnyei’s guidelines (2007), in the presentation of my results (see Section 4.1.2), I sought to describe in sufficient detail the limitations of the sample while also highlighting the characteristics that the sample shared with the target population, and, secondly, I tried to be careful about the claims I made concerning the general relevance of my findings.

With regard to Study 2, it is also important to mention that the participants were recruited online and were taking part in the study voluntarily, which could have potentially caused an overrepresentation of respondents with genuine interest in the research topic. One step I took to mitigate this problem was to make sure that the title and the objective of the questionnaire were presented to the participants “at a fairly general level” (Cohen et al., 2007, p. 57), i.e., providing enough details without being too revealing. Furthermore, I asked some of my EFL teacher colleagues who were the head of their schools’ English teams to have the questionnaire filled in during one of their meetings, thus ensuring that participants otherwise less interested in the topic are also involved. Finally, by individually sending nearly 1000 emails to English teachers across the country, I aimed to encourage participation from those who might otherwise be less motivated to join the study.

In the smaller-scale questionnaire study conducted within Phase 2 (Study 5), sampling parameters were again set up, however, it has to be noted that as in this case the study was small-scale, where generalisation needs to be done with even more caution. As was pointed out in Section 3.3.4.3, while data were collected from three different schools, and certain degree of generalisation may be justified, the primary aim of this particular questionnaire study was to explore patterns within the examined sample and to provide additional data for the classroom research for triangulation, rather than to generalise findings to a larger population.

In my qualitative studies (Studies 1, 3, and 4), I followed Lincoln and Guba’s (1985) taxonomy of quality criteria by aiming for the dependability, credibility, transferability, and confirmability of my project. As for dependability, I piloted my interview schedule for Studies 1 and 3 and my observation schedule for Study 4 and then used the same instrument (with only minimal changes) with all of my participants to ensure the consistency of data collection. I also kept an audit trail (Dörnyei, 2007), in which I recorded detailed notes of the research process, including data collection methods, data analysis decisions, and changes in the study design, parts of which are also presented in various sections of this chapter. Finally, in both the interview studies and the observation study, I used the code-recode strategy (Saldaña, 2013) by revisiting my dataset to check the consistency of my coding process.

Regarding credibility (i.e., internal validity), I subjected my interpretations and conclusions to peer debriefing by consulting regularly with my supervisor and with some of my peers in the PhD programme. I had also intended to use member checking with my participants, however, I found that within their already heavy workloads, asking them to read the transcript as well as the results section of my manuscript would have not been feasible; if any question arose during analysis, however, I got back to them and ask them to verify these portions. Concerning the classroom observation (Study 4), I anticipated that the Hawthorne effect might be at play, i.e., participants performing differently when knowing they are being studied (Dörnyei, 2007). To mitigate this effect, I made sure to introduce myself and establish rapport with the groups being studied before the observation began (Duff, 2008). While in the case of school A I had to be cautious about my pre-existing connection with the school to avoid biases, I agree with Pollard (1985) that a close familiarity with the participants and the research site may have actually encouraged more natural behaviours among the participants, which could help maintain the validity of the research.

As for transferability (i.e., external validity), when presenting my results, I sought to provide a thick description of the participants and the contexts (Dörnyei, 2007) to help the reader decide whether the outcomes are applicable to their contexts, as well. Finally, as for confirmability (i.e., objectivity) I made sure to identify my potential biases, especially in Phase 2, with regard to School A, where I had been formerly employed (see Section 3.3.1.2). I sought to maintain the integrity and validity of the research project by providing thick description of the research context, maintaining reflexivity throughout the research process, distinguishing objective data from subjective interpretations, and using peer debriefing and occasional member checking to mitigate any potential biases arising from my pre-existing connections (Dörnyei, 2007; Duff, 2008; Lundy, 2008).

Besides the quality control measures outlined above, in both Phases 1 and 2, I opted for mixing methods with the intent to ensure design validity (Teddlie & Tashakkori, 2003), i.e., to combine the qualitative and quantitative components of the project in a way that “the overall design displays complementary strengths and nonoverlapping weaknesses of the constituent methods” (Dörnyei, 2007, p. 63). In Phase 1 my main intent with mixing methods was to use an exploratory sequential design (Creswell, 2015) with Study 1 being conducted to identify some emerging themes for RQs 1 and 2 that would then inform the development of the questionnaire in Study 2, and to test how spread out the initially identified themes are in a larger population. In Phase 2, mixing methods and perspectives (that of teachers and students) served the purpose of triangulation, which helped to strengthen the credibility of this phase by “reducing the chance of systematic bias” (Dörnyei, 2007, p. 61). While convergence between the results of the sub-studies within Phase 1 and Phase 2 was deemed as the preferred outcome, divergences were also addressed, with the aim of understanding the reasons behind these discrepancies.