In this paper, we investigate joint optimization of power allocation factors (PAFs) and decoding order of users in the downlink of a non-orthogonal multiple access (NOMA) system over Nakagami- $m$ fading channels. The objective is to guarantee fairness among the users in terms of outage probability (or its complement success probability). To this end, we maximize the minimum success probability (Max-MSP) among the users when only statistical channel state information (CSI) is available at the transmitter side. We solve such a problem by first proving that at the optimal solution, all the users have a common success probability (CSP), and then proposing an efficient algorithm for finding CSP of the users. The optimal PAFs and optimal decoding order are derived in closed form based on the CSP and statistical CSI of the users. It is proved that the proposed algorithm always converges to a unique optimal solution. Furthermore, we show that in contrast to Rayleigh fading channels, in Nakagami- $m$ fading channels, the optimal decoding order not only depends on the average signal-to-noise ratio (SNR) of each user but also depends on the variance of the SNR and data rate of that user.